of the International Society for Magnetic Resonance in Medicine

British Chapter of the International Society for Magnetic Resonance in Medicine 22nd Postgraduate Symposium London 19th March 2013 Sponsors 2 Contents Welcome 4 Symposium Information 5 Practical Information 6 About CABI 6 Wellcome Collection 7 Programme 8 Abstracts – Oral Presentations Cutting-edge Techniques 15 Brain Investigations 25 Cancer, Cardiac & Nerves 37 Abstracts – Posters & Poster Pitches 43 Abstract details 70 List of Participants 71 Notes 74 Local Area Map 77 Programme 78 3 BC-ISMRM XXII Postgraduate Symposium 2013 Welcome! Welcome to London and to the 22nd annual Postgraduate Symposium of the British Chapter of the International Society for Magnetic Resonance in Medicine! You are among over 100 delegates who are joining for one day to exchange ideas, discover each others' work, and meet new members of the MR community. We hope you'll take this opportunity showcase your work, gather contacts for the future, and explore the Wellcome Collection, where this year's meeting is hosted by members of the UCL Centre for Advanced Biomedical Imaging (CABI). We are holding a slightly different symposium this year, and with your help we think it will be even more exciting than usual. Every presenter – including those with posters – will get a chance to speak in front of the whole conference. We're doing this to raise the profile of posters: the work is just as relevant, but there isn't time during a one-day meeting for 50 full talks! Our 'Poster Pitch' session (15:00 – 16:30) is your opportunity to see, in 2 minutes each, a very brief overview of the work and results from every poster presented at this year's Symposium. We still want you to seek out posters and their presenters in the full Poster Session (16:30 – 17:30), so if you're interested in following up a Poster Pitch with questions, that is the time to do it! We also hope you'll take some time to explore the excellent Wellcome Collection on the 1st floor over lunch (13:30 – 15:00), which hosts free exhibitions: currently on display are ‘Medicine Now’, a look at contemporary medical issues, and ‘Medicine Man’, a selection of items from Henry Wellcome’s own collection. In this booklet, the abstracts start on page 15. There's a full timetable of presentations on pages 9 – 13, and a programme for the day on the last page. There’s also a little about the Wellcome Collection and CABI on pages 6 & 7. If you're staying in London for the evening, we've mapped out a few recommended haunts within walking distance (penultimate page) – perhaps you’ll join us for a wind-down and drink afterwards? In any case, we hope this year’s Symposium is vibrant and fun, as well as showcasing some excellent work. Have a great time! BCISMRM Postgraduate Symposium Organising Committee at the Centre for Advanced Biomedical Imaging (CABI), UCL, 13th March 2013 4 Symposium Information Organising Committee Prof. Mark Lythgoe Nick Powell Holly Holmes Rupy Ghatrora James O’Callaghan Tom Roberts Ma Da Francesca Norris Chair Secretary Sponsorship Location Manager Abstract Coordinator Programme Website Budget Coordinator m.lythgoe@ucl.ac.uk nicholas.powell.11@ucl.ac.uk h.holmes.11@ucl.ac.uk r.ghatrora@ucl.ac.uk james.ocallaghan.10@ucl.ac.uk thomas.roberts.10@ucl.ac.uk d.ma.11@ucl.ac.uk f.norris@ucl.ac.uk Abstract Reviewers Dr. Niall Colgan Dr. Tammy Kalber Dr. Bernard Siow Dr. Daniel Stuckey Dr. Simon Walker-Samuel Dr. Jack Wells Prize Judges Posters Poster Pitches Oral Presentations Prize giving Dr. David Carmichael & Dr. Karin Shmueli Dr. Nicola Sibson & Prof. Xavier Golay Dr. Claudia Wheeler-Kingshott & Dr. Simon WalkerSamuel Prof. David Gadian Session Chairs and Speakers Cutting-edge Techniques Brain Investigations Poster Pitches Cancer, Cardiac & Nerves Prof. Steven Williams, Dept. of Neuroimaging, King’s College London Prof. Derek Jones, CUBRIC, Cardiff University Prof. Mark Lythgoe, CABI, UCL Prof. Risto Kauppinen, University of Bristol British Chapter Chair Secretary Treasurer & Membership Professor David Gadian Professor Andrew Blamire Professor Ian Marshall 5 BC-ISMRM XXII Postgraduate Symposium 2013 Practical information Conference Information Conference centre reception 020 7611 8888 Conference email cabi-ismrm2013@ucl.ac.uk WiFi Network: Wellcome Conference WiFi; Username: Conference; Password: winter13 For Wellcome Collection information, including lost property, call 020 7611 2222 or email info@wellcomecollection.org. CABI www.ucl.ac.uk/cabi Opened in 2011, the UCL Centre for Advanced Biomedical Imaging, or 'CABI', as it is affectionately known, is the first multidisciplinary research centre for experimental imaging within UCL. The collaborative approach of the Centre brings together technology and expertise from world class research groups across UCL. The Centre now hosts 11 state-ofthe-art imaging modalities: MRI, Photoacoustic Imaging, Ultrasound, Bioluminescence and Fluorescence Imaging, SPECT/CT, PET/CT/MRI, Optical Projection Tomography, Light-Sheet Imaging, Confocal Endoscopy and CT/radiotherapy. Come visit any time! CABI is located just around the corner in the Paul O'Gorman Building, 72 Huntley Street, London WC1E 6BT. 6 Wellcome Collection www.wellcomecollection.org We have been lucky to secure a spectacular venue for this year’s Postgraduate Symposium. The Wellcome Collection is just next door to UCL and CABI, and houses public exhibitions (free to visit, on the first floor) and a library (second floor). We hope you’ll take the time to explore them during the lunch break (13:30 – 15:00). Enter from Euston Road – the conference centre is downstairs. There are more toilets further downstairs. First Floor Medicine Man showcases over 500 objects and curios from Sir Henry Wellcome’s personal collection. Medicine Now is a gallery of artists’ and scientists’ responses to contemporary medical topics including genetics, disease and obesity. 7 BC-ISMRM XXII Postgraduate Symposium 2013 Programme 9:00 Williams Lounge Registration opens 9:30 Henry Wellcome Auditorium Welcome & Opening remarks Henry Wellcome Auditorium Oral presentations: Cutting-edge techniques 11:15 Williams Lounge Refreshments 11:45 Henry Wellcome Auditorium Oral presentations: Brain Investigations 9:35 13:30 Prof. Mark Lythgoe Chair: Prof. Steven Williams Chair: Prof. Derek Jones Dr. Ken Arnold (Head of Public Programmes, Wellcome Collection) will introduce the Collection just before lunch. 13:35 Williams Lounge Lunch & Exploring Wellcome Collection 15:00 Henry Wellcome Auditorium Poster Pitches 16:30 Franks & Steel Rooms Posters & Refreshments 17:30 Henry Wellcome Auditorium Oral presentations: Cancer, Cardiac & Nerves Henry Wellcome Auditorium Prizes and Concluding remarks 18:30 19:00 8 Chair: Prof. Mark Lythgoe Chair: Risto Kauppinen Prof. David Gadian and Prof. Mark Lythgoe Wellcome Collection closes Oral presentations: Cutting-edge Techniques Chair: Prof. Steven Williams O10 Alessandro Proverbio Multimodality investigation of microstructures by the combination of diffusion NMR and diffuse optical spectroscopy O11 Eleanor Evans MRI and PET image-derived vascular input functions for quantitative kinetic modelling in mice O12 Tom Roberts Multi-Parametric Magnetic Resonance Imaging of Myocardial Infarction in Mice at 9.4T O13 Frank Riemer Density compensation function optimization for 23Na-MRI reconstruction from non-Cartesian trajectories O14 Hannah Hare Carbogen for CVR? O15 James Ross Rapid Field-Cycling MRI using Fast Spin-Echo O16 James Meakin VS-TILT: Velocity Selective Arterial Spin Labeling without diffusion effects O17 Miguel Goncalves Investigating systemic and tumour-specific fluctuations in tumour R2* measurement with independent component analysis and pulse oximetry O18 Matthew Rowe ND-Track: Tractography utilising parametric models of white matter fibre dispersion 9 BC-ISMRM XXII Postgraduate Symposium 2013 Oral presentations: Brain Investigations Chair: Prof. Derek Jones O20 Alan Stone Optimisation of calibrated FMRI for the detection of regional alterations in absolute CMRO2 O21 Simon Richardson Viable and Fixed White Matter: DTI and Microstructural Comparisons at Physiological Temperature O22 Esther Warnert Measuring Tissue Perfusion in the Human Brainstem in Normo- and Hypercapnia Using Multi Inversion Time Pulsed Arterial Spin Labelling O23 Francesco Grussu Probing Spinal Cord Microstructure with the Neurite Orientation Dispersion and Density Imaging O24 Gavin Winston Clinical utility of NODDI in assessing patients with epilepsy due to focal cortical dysplasia O25 Jessica Steventon Environmental Enrichment retards the development of neuropathology and ameliorates motor deficits in a mouse model of Huntington’s Disease: an in vivo MRI study O26 Marilena Rega Imaging pH changes in piglet brain after acute hypoxia-ischemia using Amide Proton Transfer (APT) O27 Jack Wells Novel Biomarkers of Tau pathology in a mouse model of Alzheimer’s Disease: CEST, ASL and Glucose-CEST O28 Richard Joules Connectivity effects of Ketamine its modulation by Risperidone and Lamotrigine O29 Muhammad Chowdhury A novel method of minimizing EEG artefacts during simultaneous fMRI: a simulation study 10 Poster Pitches Chair: Prof. Mark Lythgoe P1 Andrada Ianus Design of trapezoidal oscillating gradients for diffusion MRI P2 Andreas Glatz Novel, unsupervised method for iron deposit segmentation in the basal ganglia P3 Anita Banerji The utility of synthetic data for quantitative assessment of a DCE-MRI registration algorithm P4 Ben Duffy MRI of epilepsy therapy: Dexamethasone may exacerbate cerebral oedema in a rat model of status epilepticus P5 Christopher Parker Agreement in Reproducibility of Whole-brain Structural Connectivity Networks with Alternative Pipelines P6 Dimitra Flouri Fitting the two-compartment filtration model in renal DCE-MRI by linear inversion P7 Eleanor Berry Optimised encoding scheme for vessel-selective arterial spin labelling P8 Grzegorz Kowalik Assessment of cardiac timing intervals using high temporal resolution realtime spiral phase contrast with UNFOLD-SENSE P9 Henrietta Howells The role of fronto-parietal networks in mental imagery P10 Holly Holmes Tensor-based morphometry as a sensitive biomarker of Alzheimer’s disease neuropathology in a Tau transgenic mouse (rTg4510) P11 Ilona Lipp Low repeatability of BOLD during emotion processing is not due to physiological noise P12 James O'Callaghan In vivo Diffusion Tensor Imaging is sensitive to microstructural changes in both white and grey matter in the TG4510 mouse model of Alzheimer’s disease P13 Konstantinos Papoutsis Construction of a 4-Channel Transmit Neck Array for pCASL Tagging at 7 Tesla and Comparison with a Head Coil P14 Ma Da Cortical thickness map: an automatic quantification of cerebral cortex for in vivo mouse brain MRI P15 Mark Mikkelsen Comparison of the Reproducibility of a GABA-Edited Magnetic Resonance Spectroscopy Technique with and without Macromolecule Suppression 11 BC-ISMRM XXII Postgraduate Symposium 2013 P16 Neil Jerome Improved Confidence in Diffusion Metrics from ‘Post-Navigator’ Registration of Individual Coronal Signal Average Images in Abdominal Diffusion-Weighted MRI P17 Nick Powell Automated high-throughput morphometric phenotyping of mouse brains and embryos P18 Patxi Torrealdea GlucoCEST for the detection of human xenograft glioblastoma at early stage P19 Pete Lally The Impact of Group-wise Diffusion Tensor Registration on Tract-Based Spatial Statistical Analysis of White Matter Microstructure in Neonatal Encephalopathy P20 Rajiv Ramasawmy Multi-Slice Look-Locker FAIR for Hepatic Arterial Spin Labelling P21 Robin Simpson Measuring myocardial velocities with high resolution using retrogated spiral phase velocity mapping P22 Ruth Oliver Comparison of Bayesian and Linear Regression-based Partial Volume Correction in Single Time Point ASL P23 David Peat High polarization of nuclear spins mediated by nanoparticles at millikelvin temperatures P24 Suejen Perani Intracranial (icEEG)-fMRI: mapping brain networks associated with alpha and beta in sensorimotor cortex. P25 Tingting Wang The Influence of Macroscopic and Microscopic Fibre Orientation Dispersion on Diffusion MR Measurements: a Monte-Carlo Simulation Study P26 Valeria Parlatini Meta-Analysis of the Functional Correlates of Fronto-Parietal Networks 12 Oral presentations: Cancer, Cardiac & Nerves Chair: Risto Kauppinen O30 Emily Wholey Lactate and glutamine as potential MRS-detectable metabolic biomarkers of treatment response to the novel HSP90 inhibitor NVP-AUY922 in human breast cancer cells O31 Evangelia Kaza Diffusion Weighted MR imaging using an Active Breathing Coordinator to support Radiotherapy treatment planning O32 Guido Buonincontri Multi-modal MRI and FDG-PET for assessment of treatment in the infarcted mouse heart O33 Sean Johnson Comparison of Arterial Spin Labelling and R2* as Predictive Response Biomarkers for Vascular Targeting Agents in Liver Metastases O34 Zaid Mahbub Measurement of magnetization transfer effects in the Brachial Plexus: comparison with T2 and Diffusion effects 13 BC-ISMRM XXII Postgraduate Symposium 2013 14 Abstracts: Oral Presentations Cutting-edge Techniques 15 Multimodality investigation of microstructures by the combination of diffusion NMR and diffuse optical spectroscopy Alessandro Proverbio1, Bernard Siow2, Adam Gibson1, Daniel Alexander3 1 Department of Medical Physics and Bioengineering, UCL, WC1E 6BT London; 2 Centre for Advanced Biological Imaging, UCL, WC1E 6DD, London; 3 Centre for Medical Image Computing and Department of Computer Science, UCL, WC1E 6BT London Introduction Investigation of tissue microstructure with non-invasive histology is a key challenge for medical imaging. Diffusion NMR (dNMR) can investigate the dimension of compartments (such as cells or axons) restricting and hindering the diffusion of water molecules. Diffuse Optical Spectroscopy (DOS) measures the Temporal Point Spread Function (TPSF) of the photons emerging from a sample, and is sensitive to size and density of scatterers in the tissue (e.g. nuclei and organelles). The two techniques may complement each other, yet no demonstration of a combined model exists in literature. Here, we provide a proof of concept that combining information from dNMR and DOS, via a joint signal model, improves estimation of microstructural features compared to biophysical model of tissue microstructure informed by either modality alone. Methods To demonstrate the idea, we used an oil-in-water emulsion sample (Sainsbury's commercial light mayonnaise) containing oil droplets in water and emulsifiers. The microstructure in the sample is modeled as a set of spherical elements with average radius r, log-normally distributed with spread parameter σ, and volume fraction ψ. The diffusion of oil molecules is constrained inside the droplets while the interface between oil and water causes optical scattering. dNMR measurements were performed with a Stimulated Echo sequence applied with a 9.4T Varian experimental system (∆=100-700ms, δ=3ms G=0-0.95 T/m, TR=4s, with minimum echo time). We acquired 4 repetitions along each direction of three orthogonal gradient directions for all the 30 combinations of parameters. The signal model assumes restricted diffusion in spherical Figure 1: The combined model compartments. The parameters are diffusivity coefficient D, r and σ . DOS measurements were performed with a time-domain system transilluminating a sample in a container of 17x48x52mm 3 across the smaller dimension with a 780nm wavelength pulse laser, and a detector measures the TPSF. The model assumes a Diffusion Approximation, and uses Mie theory to relate a single scatterer size parameter, r, and volume fraction, ψ, to the TPSF. Table 1: Estimation of r (and its standard deviation in brackets) from 150 synthetic Additional parameters are the apparent scattering and absorption coefficients µs' datasets generated with SNR=16 and and µa introduced respectively. r=1,2,3µm, σ=1.5µm, and ψ=0.2. The combined model and estimation procedure are presented in figure 1. Signals are also fitted for each modality individually by minimizing the sum of squared differences of the model and measured signal. DOS alone provides r and ψ, dNMR alone provides r and σ. Combined model fitting minimizes a weighted sum of the fitting errors from the two signals to obtain, r, σ, and ψ, at the same time. Confocal Laser Scanning Microscopy (CLSM) provides ground truth r, σ, and ψ for the sample. For further validation, synthetic datasets were independently generated with different SNR at realistic values of parameters by introducing a Figure 2: Estimation of r and ψ obtained Rician noise in dNMR signals and a Gaussian noise in DOS. from the experimental signals and their Results Table 1 shows recovered values of r from the synthetic data from dNMR accuracy obtained with MCMC. alone, DOS alone, and the combined model (Com). The combined model provides more accurate and precise parameter estimates especially for larger values of r. In dNMR, a low diffusivity coefficient (10-11µm2/s) and a consequently long ∆ may introduce a bias for the estimation of larger r. DOS has a better performance with smaller values of r, accordingly to the reduction of accuracy observed when Mie theory is adopted for the sizing of scatterers with a dimension much larger then the laser wavelength. Figure 2 shows the estimates of r and ψ obtained from experimental signals. The combined model shows much more accurate and precise estimates of r compared to dNMR alone and slight improvements over dNMR alone. The error bars represent the standard deviation of the chain of estimates obtained with a MCMC algorithm. The standard deviation in combined model is 1/4 of dNMR ones. Finally, combining dNMR with DOS improves the estimation of ψ reducing the error to less then 2% even though dNMR is not directly sensitive to ψ; the improvement comes about by improving the estimate of r, which is linked to ψ via DOS model. Discussion and conclusions A common model informed by both DOS and dNMR signals refines the estimation of parameters detectable with both of them since it fuses complementary information. Moreover, the fitting of the noncommon parameters is facilitated, leading to a better performance. The two modalities exploit different physical phenomena thus providing complementary information. In conclusion, a model informed by the two modalities leads to a natural enhancement of the parameters estimation, since the model exploits the strength of DOS and dNMR. A natural application of this modality is the study of cellular structure in cancer, where DOS and diffusion MRI can be combined. MRI and PET image-derived vascular input functions for quantitative kinetic modelling in mice 1 1 1 1 1,2 E. Evans , A.O. Ward , G. Buonincontri , T. A. Carpenter and S. J. Sawiak 1 Wolfson Brain Imaging Centre, University of Cambridge, Box 65 Addenbrooke’s Hospital, Cambridge, UK, CB2 0QQ 2 Behavioural and Clinical Neuroscience Institute, University of Cambridge, Downing Street, Cambridge, UK Background and Aims Positron emission tomography (PET) can be used for quantitative, specific molecular imaging. The most useful parameters come from compartmental modelling of tissue signals, however this requires accurate levels of blood plasma tracer concentrations in particular the arterial input function (AIF). In human scans this is achieved with blood sampling at regular intervals. In small animals however, this is prohibitive due to small total blood levels (rats: ~20ml; mice: ~2ml [1-2]). It is possible to derive AIFs from the images if appropriate voxels can be selected, though the low temporal and spatial resolution of PET make this difficult. MRI offers much greater spatial and temporal resolution (e.g. x10) and with combined PET-MR it could be possible to avoid blood sampling altogether. Here, in a preliminary experiment, we compare image-derived AIFs using PET and MRI images acquired sequentially. Materials and Methods MRI EPI scans were performed on a 4.7T Bruker BioSpec system (TR/TE 250/9ms, 2 spatial resolution 110×200μm , 5 slices, thickness 1.5mm, 400 frames) starting to include bolus injection of 0.3mM/kg Gd-DTPA through the cannulated tail vein. PET Following MRI, the animal bed was transferred to the Cambridge split-magnet 18 PET/MR system [3] where 30 MBq F-FDG was administered via the same cannula used for MRI. Data were acquired in listmode for 1 hour for later reconstruction (2D FBP framed as 30×1s, 30×5s, 12×10s and 11×300s). MRI AIF, a progressive voxel inclusion scheme was applied to a rectangular ROI which encompassed the posterior cerebral artery (PCA). Voxel inclusion thresholds were th th empirically determined with starting values of >90 percentile peak height, >10 th th percentile area under the curve, <50 percentile first moment and <50 percentile FWHM based on schemes in the literature [4-5]. MR AIFs were calculated as the mean signal from 5 voxels with the largest signal heights which survived all thresholds, a typical example is shown in Figure 1. PET AIF A seeded growth region in the left ventricle (LV) of the heart was used across the time course (~50 voxels). Results and Discussion Peak heights of MR and PET AIFs were normalized to 1 and flush curve peaks aligned for each animal. We saw significant variability in the injection consistency between animals (Figure 2a). Fits of gamma variate functions to subject AIFs confirmed that PET and MR curves exhibit similar shapes in the first pass (mean rise times: PET=4±1s, MR=2.8±0.5s). PET AIFs also maintained a higher activity at 90s (PET = 0.5±0.2, MR = 0.04±0.05, fraction mean peak height ± SD). Reducing the MRI resolution to that of the PET data gave a similar curve to the PET AIF (Figure 2c). This suggests that partial volume effects can explain the discrepancy seen between the functions. Figure 1 EPI image at peak of first passage of bolus, with automatically selected arterial voxels for AIF calculation shown in orange Figure 2 (a) Variability of PET AIFs for all subjects, (b) Single subject PET and MR AIF comparison, (c) MR data rebinned into PET timing resolution before normalization. Error bars not shown for clarity (a) (b) Conclusions We have shown that the MRI AIF can be determined automatically giving results similar to the PET AIF. Partial volume effects make mouse input functions difficult to derive with PET imaging alone. These preliminary data suggest that automatically obtained MRI-based AIFs are feasible in combined PET/MR. The next step will be simultaneous measurement of AIFs with PET, MRI and blood sampling for three-way cross-validation. References [1] Lee and Blaufox, J Nucl Med, 25, 1, 72–76, (1985), [2] Riches et al., J. Physiol, 228, 2, 279–84, (1973), [3] Lucas et al., Technol Cancer Res Treat 5(4):337-41 (2006) [4] Singh et al., JMRI, 29, 1, 166–176, (2009), [5] Peruzzo et al. , Comput. Meth. Prog. Bio, 104, 3, e148–e157, (2011) (c) MULTI-PARAMETRIC MAGNETIC RESONANCE IMAGING OF MYOCARDIAL INFARCTION IN MICE AT 9.4T 1,2 1 3 4 1 *T.A. Roberts , *A.E. Campbell-Washburn , R.K. Dongworth , D.L. Thomas , A.N. Price , 3 5 3 1 D.M. Yellon , P.J. Scambler , D.J. Hausenloy , M.F. Lythgoe . 1 Centre for Advanced Biomedical Imaging, University College London, UK. 2 Centre for Mathematics and Physics in the Life Sciences & Experimental Biology, UCL, UK. 3 The Hatter Cardiovascular Institute, University College London, UK. 4 Dept of Brain Repair and Rehabilitation, UCL Institute of Neurology, UK. 5 Institute of Child Health, University College London, UK. Introduction: Magnetic resonance imaging (MRI) is widely used for measuring cardiac function and quantifying infarct size following myocardial infarction in both clinical and basic science settings. The utility of MRI can be 1 2 extended to additionally measure myocardial oedema and perfusion deficits in vivo. Oedema and perfusion represent potentially important endpoints for determining long-term prognosis following myocardial infarction, however the in vivo measurement of these parameters is not well characterised to-date. We have developed an in vivo, multi-parametric MRI assessment, encompassing cardiac function, infarct size, oedema and myocardial perfusion measurements, which can be used for a thorough in vivo characterization of myocardial tissue following infarction. In this study we demonstrate the applicability of this MRI platform in reperfused and non-reperfused myocardial infarcts. Methods: Animal Preparation: Mice were subjected to open chest surgery to initiate myocardial ischaemia by occlusion of the left anterior descending artery (LAD). Non-reperfused mice received permanent ligation of the LAD whereas reperfused mice received LAD occlusion for 30 minutes followed by LAD reperfusion. MRI: Mice were scanned 72 hours following this procedure using a 9.4T Agilent (Santa Clara, US) scanner with a 35mm volume coil (Rapid MR International, Rimpar, Germany). Four MRI techniques were used: CINE for 3 4 cardiac function (not shown), fast spin echo-based T2 mapping for oedema , multi-slice ASL for perfusion and 5 late gadolinium enhancement for visualization of infarction . Total scan time was approximately 2.5 hours per animal. Results: Example MRI data from mid-ventricular slices of a reperfused and non-reperfused mouse are presented (Figure 1). LGE images (i) show clear delineation of infarction (areas of myocardial hyperintensity). As expected, the extent of infarction is greater in the non-reperfused animal. T2 maps (ii) show well-defined regions of elevated T2 signal (areas of red) which are considered to indicate regions of myocardial oedema. The increased uniformity of T2 signal elevation in the reperfused mouse is likely due to the reperfusion of damaged vasculature which does not occur in the non-reperfused model. ASL perfusion maps (iii) of the non-reperfused myocardium demonstrate a clear perfusion deficit (dark blue) as expected. Interestingly, there also appears to Figure 1: Multi-parametric MRI of reperfused and non-reperfused myocardium. (i) LGE shows infarct, (ii) T2 maps show oedema be a subtle perfusion deficit in the reperfused and (iii) ASL shows perfusion. myocardium. Conclusions: This pilot study demonstrates the T2 mapping and ASL sequences developed here provide a good depiction of myocardial oedema and perfusion deficit. Together, this multi-modal MRI platform may be used as an investigative tool for a more in-depth assessment of pathophysiological changes in the in vivo myocardium following infarction. [1] Giri, Shivraman, et al. Journal of Cardiovascular Magnetic Resonance (2009). [2] Vandsburger, Moriel H., et al. Magnetic Resonance in Medicine (2010). [3] Bohl, Steffen, et al. 13th Annual SCMR Scientific Sessions (2010). [4] Campbell‐Washburn, Adrienne E., et al. Magnetic Resonance in Medicine (2012). [5] Price, Anthony N., et al. Journal of Cardiovascular Magnetic Resonance (2011). Carbogen for CVR? a Hannah Harea, Michael Germuskaa, Daniel Bultea FMRIB Centre, Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, UK Background & Aims Measurement of cerebrovascular reactivity (CVR) can give valuable information about existing pathology and the risk of adverse events such as stroke. Regional values of CVR are typically acquired by measuring the flow response to CO2 enriched air using arterial spin labelling (ASL) or blood oxygen level dependent (BOLD) imaging. Provided all other variables (PaO2, CMRO2) remain approximately constant, as is assumed with an air/CO2 stimulus, either imaging technique can be used to measure CBF correlated signal changes and hence estimate CVR. Recently several studies [1,2] have used carbogen gas (composed of CO2 with balance oxygen) as an alternative stimulus, implicitly assuming that increasing the oxygen fraction of inspired air will have no effect on CVR measurement. We tested this hypothesis by acquiring values for CVR using ASL and BOLD data taken during a single scan with stimuli of (a) 5% CO2 in air and (b) 5% CO2 in oxygen (carbogen-5). Methods 9 healthy volunteers were recruited (mean age 25±4 years, 2 female), and scanned on a 3T Siemens Verio scanner using a 32-channel receive-only head coil. Subjects wore a close fitting gas mask over the mouth and nose, through which we implemented the following paradigm: 30s breathing medical air, then 1 min 5% CO2 in air; 1 min carbogen-5; 1 min 5% CO2 in air; 1 min carbogen-5; where each 1 min block was followed by 1 min of medical air. Gas delivery was 25L/min, and a sampling tube was used to measure the gas composition in the mask using oxygen and CO2 analysers. A single TI dual echo pulse sequence was used to acquire data for both ASL and BOLD following a single RF excitation. Using a pseudo-continuous ASL scheme [3] and an EPI readout (TR=4s, TE(1)=16ms, TE(2)=35ms) 23 slices (each 64×64 voxels, voxel size 3×3×5mm or 3.4×3.4×5mm to ensure full brain coverage) were acquired for each subject, where slice thickness was 5mm with inter-slice gap of 0.5mm. Age Mean ± SD 24 ± 3 CVR for CO2 in air (% change per mmHg) ASL BOLD 4.46 ± 1.80 0.11 ± 0.03 CVR for carbogen-5 (% change per mmHg) ASL BOLD 4.97 ± 1.30 0.36 ± 0.06 BOLD CVR 0.5 0.4 0.3 0.2 0.1 Carbogen-5 CO2 in air 0.0 0 2 4 6 8 10 12 Flow CVR Comparison of CVR values as measured by flow (ASL) and BOLD response, with linear models fitted. For 5% CO2 in air R2=0.33, p=0.02; for carbogen-5 R2<0.001, p=0.99 (excluding outlier hemispheres marked “×”). Flow CVR 12 10 8 Carbogen-5 Results & Analysis Most subjects responded well to the protocol. The data collected from one subject, who requested the scan be stopped near the end of the final carbogen block, was classified as an outlier and was thus excluded from all subsequent analysis. No subjects exhibited significant changes in breathing rate between the 3 gas mixtures. Values for end tidal partial pressure of CO2 (PETCO2) were extracted from respiratory data as an average of the last 3 breaths taken during each stimulus block; average baseline PETCO2 was visually determined for each subject. PETCO2 was used to deduce the change in PCO2 in units of mmHg [4]; 5% CO2 in air increased PETCO2 by 12.1±2.2mmHg, carbogen-5 by 10.0±2.5mmHg. The FEAT tool in FSL (FMRIB, Oxford, UK) was used to extract % increases in CBF for each brain hemisphere (from absolute ASL signal, taking tag/control signal modulation into account) and BOLD signal during stimuli, which were used to calculate CVR values, as summarised in the table below and the graphs on the right. Note that for ASL data taken during a carbogen stimulus it was necessary to correct for the change in T1 of the blood compared to baseline due to increased PaO2 [5]. 0.6 6 4 2 0 0 2 4 6 CO2 in air (b) 8 10 Comparison of flow CVR values during stimuli of CO2 in air versus carbogen-5, with linear model fitted. R2=0.66, p<0.001 (excluding outlier). Discussion & Conclusions Our key findings were that 1) CVR as measured by BOLD and ASL response appear to be correlated for CO2 in air (in agreement with previous findings [6]) but not when using carbogen as a stimulus and 2) intrasubject flow CVR as measured by ASL is affected by PO2 and is not consistent between the two gas stimuli. Because of the high oxygen content of carbogen-5 gas it is not possible to identify how much of an ensuing increase in BOLD signal is due to increased blood flow (as opposed to simply an increase in venous oxygen saturation), so CVR cannot be accurately determined. Therefore, BOLD imaging should not be used with a carbogen stimulus to determine CVR. We also found that although flow CVR values (as measured by ASL) to air/CO2 versus carbogen are correlated, the values obtained are not directly comparable. A further potential confound of using carbogen gas is that the interplay between the Bohr and Haldane effects is likely to influence diffusion gradients in the lungs affecting the process of gas exchange and partial pressure relationships in the lungs, blood, and tissues. Under these conditions changes in PETCO2 measures may no longer be accurate indicators of arterial PCO2 changes. 1. Cantin S, et al. Neuroimage 2011:(2):579-587. 2. Hamzei F, et al. Neuroimage 2003:(2):1393-1399. 3. Dai W, et al. Magn Reson Med 2008:(6):1488-1497. 4. Young WL, et al. JCBFM 1991:(6):1031-1035. 5. Bulte DP, et al. JCBFM 2007:(1):69-75. 6. Mandell DM, et al. Stroke 2008:(7):2021-2028. Rapid Field-Cycling MRI using Fast Spin-Echo P. James Ross, Lionel M. Broche, David J. Lurie Aberdeen Biomedical Imaging Centre, University of Aberdeen, AB25 2ZD, Scotland, UK Introduction Fast Field-Cycling MRI (FFC-MRI)1 is an emerging technique that adds a new dimension to conventional MRI by making it possible to rapidly vary B0 during a pulse sequence. By doing this it is possible to observe how the NMR relaxation rates of biological tissue vary with magnetic field strength. Conventional relaxometric imaging is limited by lengthy scan times, since to estimate R1 at least two images (e.g. IR and SR) must be acquired at each field strength. In this abstract we describe an adaptation of the well known Fast Spin-Echo pulse sequence3 for FFC-MRI, named Field-Cycling Fast Spin Echo (FC-FSE) which enables relaxometric imaging in a fraction of the time that would otherwise be required. Methods Figure 1: Dispersion curves for a phantom of Imaging was carried out on a home-built whole-body field-cycling imager with a 59 cross-linked BSA obtained using the FC-FSE 4 mT detection field . The system uses a commercial console (SMIS Ltd., U.K.). sequence (solid dots) show good agreement with For each experiment a saturation recovery and inversion recovery image are acquired at results from a commercial relaxometer (open the detection field. A field-cycling inversion recovery image is then acquired for every circles). evolution field of interest. During the inversion recovery period B0 is rapidly switched to a different field for an evolution period, which is typically of the order of T1. The field is then switched back to the detection field and the image is acquired. R1 is estimated at each field using a two-point method. Relaxometry was also performed on small samples using a commercial bench-top field-cycling relaxometer (SMARtracer, Stelar s.r.l., Italy). Results Figure 1 shows good agreement between the R1 dispersion results obtained using the FCFSE sequence and those obtained using the relaxometer for a phantom consisting of crosslinked bovine serum albumin (BSA). FC-FSE images from a volunteer’s thighs using an ETL of 4 (Figure 2) exhibit virtually no artifacts from field-instability. A dispersion curve (Figure 3) obtained from the outlined region-of-interest in muscle shows enhanced relaxation at specific frequencies, known as quadrupole peaks, arising due to 1H-14N crossrelaxation in immobile protein molecules within the muscle. The total scan time was ~30 minutes compared to the 4 hours that would have been required using conventional relaxometric imaging. Conclusions This work has demonstrated that relaxometric imaging can be performed up to 8 times faster relative to the basic procedure, with virtually no sacrifice in the accuracy of R1 determination. This paves the way for clinical relaxometric studies with acceptable scan times. Figure 2: Image of a volunteer’s thighs obtained using the FC-FSE sequence with a speed up factor of 4. RoI delineates muscle, from which a dispersion curve was obtained. Acknowledgements The author acknowledges funding from the EPSRC through the Centre for Doctoral Training in Integrated Magnetic Resonance. References [1] Lurie, D.J. et al., C.R. Physique 11, 136-148, 2010. [2] Pine, K.J. et al., Magn Reson Med, 63, 1698-1702, 2010. [3] Hennig, J. et al., Magn Reson Med, 3, 823-833, 1986. [4] Lurie, D.J. et al., Phys Med Biol, 43, 1877-1886, 1998. Figure 3: R1 dispersion curve for the RoI shown in Figure 2. The quadrupole peaks arising due to immobile proteins in muscle are clearly visible (see arrows). VS-TILT: Velocity Selective Arterial Spin Labeling without diffusion effects James A. Meakin1,2, Natalie L. Voets1 and Peter Jezzard1 1 FMRIB Centre and 2Department of Oncology, University of Oxford Introduction: Velocity Selective Arterial Spin Labeling (VS-ASL) is a pulsed ASL technique that is insensitive to bolus arrival time by imaging spins that have decelerated from above to below a cutoff velocity (vc), which is typically set to the velocity of small arterioles (2 cm/s) [1]. In the tag condition a Velocity Selective (VS) preparation saturates spins in laminar vessels by playing out a bipolar gradient with first moment m1 = π/(γvc) (fig 1a). In the control condition these gradients are not played out so that m1 = 0 (fig 1a). Along with m1 being different between tag and control scans (which generates the velocity selective contrast) there is also a difference in diffusion b-value. This b-value difference causes diffusive tissue spins to erroneously appear as positive signal in the ΔM subtraction and an overestimation of perfusion (f), which is especially Figure 1. VSASL (a) and VSASL-TILT (b) VS preps for vc = 2cm/s problematic in areas of high ADC such as CSF or edema [2,3]. In this with absolute RF (top), tag gradient scheme (middle) and control work we design a velocity selective prep that has the same b-value in both tag and control, and determine the magnitude of the diffusion error gradient scheme (bottom). in healthy grey matter. Methods: Diffusion balanced VS pulse design: A BIR-8 VS prep [4] (fig. 1a) is played out twice (VS1 and VS2, fig. 1b). In the tag condition VS1 and VS2 have the same gradient polarity so the total velocity weighting is twice the m1 of an individual VS unit. In the control condition we reverse the gradient polarity of VS1 so that the m1 from VS1 is negated by VS2, but the b-value is identical between tag and control. This is conceptually similar to TILT, a pulsed ASL labelling technique [5] where two +90o RF pulses invert blood in the neck, and +90o-90o RF create the 0o control to balance MT effects. Measurements of tagging efficiency in a flow phantom: A tube filled with tap water was connected to a peristaltic pump and the average velocity increased from 0 to ± 10 cm/s. VSASL and VS-TILT with vc = 2 cm/s were applied prior to a flyback EPI readout with TR = 10 s, TE = 20 ms, 4 tag control pairs per velocity. The saturation efficiency was defined voxelwise as α(v) = (Mc–Mt)/M0(v), where M0(v) is the signal at water velocity v without a VS prep. Phantom and in-vivo ΔM variation with vc: As vc is reduced to probe smaller arterioles both m1 and the b-value difference in VSASL increase. The signal difference in the VSASL ΔM subtraction due to tissue diffusion is then given by � � � � � � �� ∆M = M0 · exp −TE T2 · exp −TI T1 · 1 − exp −(TR−TI) T1 Figure 2. Above: Mean ± SD saturation efficiency in flowing water with vc = 2 cm/s. Right: Control (top) and tag (bottom, windowed x10) images for VSASL & VS-TILT at flow v = 10 cm/s. · (1 − exp(−b · ADC)) · α where ADC is the apparent diffusion coefficient of the voxel and b is the b-value difference between the tag and control VS prep. VSASL and VS-TILT was performed in a doped water phantom (no perfusion) and two healthy volunteers with vc = 1, 2, 4, 8 & 16 cm/s (corresponding VSASL b-value = 5.32 to 0.03 s/mm2), 16 tag-control pairs, TR = 5 s, TE = 30 ms, 7 slices, 3.6 × 3.6 × 8 mm voxels, crushed SE-EPI readout. In the doped water phantom the VS preps were applied with inflow time TI = 10 ms. In the healthy volunteers TI = 1 s, and an additional set of experiments were performed where the tagging efficiency for VSASL was matched to the VS-TILT by playing out two VS preps to give the same m1 in the tag, but for the control condition the gradients were turned off. Perfusion was quantified using a single compartment model [4]. Results: Figure 2 shows that the VS-TILT prep is producing an efficient saturation of flowing spins (88.2±2.4%), which is comparable to the VSASL prep (89.2±2.3%) (fig. 2). Artefacts in a static phantom due to spin diffusion in VSASL (fig. 3a) are almost eliminated by VS-TILT (fig. 3b), which matches our predictions given the known phantom T1, T2 and ADC (fig. 3c). In vivo the apparent VSASL perfusion increases as vc reduces (fig. 4d), in line with other studies [1]. However, in VS-TILT the perfusion decreases with vc, and the average perfusion is lower than VSASL even when the tagging efficiency is matched. Figure 3. VSASL (a) and VS-TILT (b) mean control – tag subtraction in a static phantom. Mean subtraction and model prediction of the error given the known ADC, T1 & T2 (c). Figure 4. Perfusion images for subject 1 at different vc for VSASL (a), VSASL with matched tagging efficiency to VS-TILT (b), VS-TILT (c) and mean signal in the Grey Matter mask for the three preparations (d). Discussion: We have demonstrated that spin diffusion can result in hyperperfusion signal in VSASL (fig 3a). We have developed a VS-TILT labelling scheme that is diffusion balanced (fig 1b) and produces a comparable labelling efficiency for flowing spins (fig 2). The loss of perfusion contrast in grey matter from VSASL to VS-TILT is not due to labelling efficiency differences as this is matched, so must be due to the difference in b-value between tag and control. However, the magnitude of the VSASL signal loss is not predicted using literature grey matter ADC (0.7 mm2/s) nor CSF partial volume. We hypothesise that the loss of signal would be explained by large ADC values measured when applying low b-values (0 to 200 s/mm2) due to pseudo-diffusion effects, or Intra Voxel Incoherent Motion [6]. Conclusions: We have introduced a Velocity Selective labelling scheme, VS-TILT, which is insensitive to diffusion. We used this to demonstrate that a significant proportion of the VSASL signal is not due to decelerating spins, but is instead due to diffusion effects. References: [1] Wong et al. MRM 55:1334 (2006) [2] Guo et al. proc. ISMRM p2116 (2011) [3] Meakin et al. proc. ISMRM p.2148 (2013) [4] Meakin & Jezzard MRM 69:832 (2013) [5] Golay et al. JMRI 9:454 (1999) [6] Federau et al. Radiology 265:874 (2012) Investigating systemic and tumour-specific fluctuations in tumour R2* measurement with independent component analysis and pulse oximetry M.R. Gonçalves1*, S. Walker-Samuel1*, S.P. Johnson2, R.B. Pedley2, M.F. Lythgoe1 1 UCL Centre for Advanced Biomedical Imaging, Division of Medicine and Institute of Child Health, University College London, UK 2 * UCL Cancer Institute, London, UK joint first authors Target audience: Researchers interested in cancer studies, particularly cycling tumour hypoxia. Also, Independent Component Analysis (ICA) applications. Introduction: Using blood oxygen level dependent (BOLD) MRI, solid tumours have previously been found to exhibit transient patterns of hypoxia with subsequent reoxygenation and/or reperfusion [1, 2], thought to be caused by vascular instability and raised interstitial fluid pressure [3, 4]. However, the influence of systemic changes in blood oxygenation on these cyclical events has not been investigated. Independent component analysis (ICA) [5] has been used in the brain to identify resting state network patterns of activation [6], by detecting individual features in coherent patterns of oscillations in blood flow and/or oxygenation. As tumour ‘resting state’ R2* fluctuations are likely to be composed of a combination of systemic and tumour-specific effects, we present here a method based on ICA to characterise both contributions. Methods: Tumour models and MRI: 5x106 SW1222 (n=6) or LS174T (n=5) colorectal carcinoma cells were injected subcutaneously into MF1 nu/nu mice. Two weeks after injection, tumours reached an approximate volume of 500mm3, and were then imaged on a 9.4T Agilent VNMRS 20cm horizontal-bore system, with a 39mm birdcage coil, using a multi-slice, spoiled multi-gradient echo (GEMS) sequence. Mice were anaesthetised using isoflurane (1.25% in medical air). Respiratory frequency varied between 43-92 breaths/min and temperature was maintained at 36.7 °C. Dynamic, multi-slice gradient echo data were acquired for 60min in each tumour, from which R2* was calculated. Sequence parameters included: TR=59.62ms, 5 echoes, TE1=2ms, echo spacing=2ms, 5 slices, 64x64 matrix, voxel volume 312x312x1500μm, flip angle=20°. Arterial hemoglobin O2 saturation (O2sat) was simultaneously measured on the thigh, by pulse oximetry (MouseOx®, Starr Life Sciences Corp., Oakmont, PA). Post-processing: Voxel-wise post processing was performed in Matlab. Maps of the standard deviation of resting-state R2* timecourses (RS(SD)) were produced for each tumour to identify regions undergoing fluctuations in oxygen saturation and/or perfusion. ICA maps were created from decomposition of R2* timecourses into independent components (ICs) with ICA [5]. After decomposition, ICA temporal patterns were correlated with the systemic O2 saturation timecourse and categorised into systemic (p < 0.01) or tumour-specific (p > 0.01) ICs. Z-score maps were produced from ICA spatial maps, one for each IC, and a zscore threshold of 2.2 was imposed to identify voxels whose R2* timecourse has similarities with a given IC. Finally, the ICA maps were averaged across ICs into a single map, considering only the z-scores that survived the threshold. Results: An example RS(SD) map from an SW1222 tumour is shown in Fig.1a, which shows regions with raised temporal fluctuations (blue arrows). Maps of ICA zscores are also shown, from the same tumour, categorised into components that either significantly correlate with systemic changes in O2sat (Fig. 1b), or show no correlation with systemic O2sat (Fig.1c). In this example, the regions identified with raised RS(SD) were not associated with systemic changes in O2sat, rather were specific to the tumour. Plots of the mean systemic and tumour-specific ICA timecourses from these data are shown in Figs. 1d and 1e, which show clear differences in temporal dynamics. Across the cohort of tumours, the number of pixels with a tumour-specific or systemic mean z-score greater than 2.2 was compared between tumour types (Fig. 2). According to this analysis, the mean percentage of tumour voxels displaying tumour-specific fluctuations was 86.0% and 96.0% for SW1222 and LS174T tumours, respectively (p=0.13, Mann-Whitney U test); the mean percentage of tumour voxels displaying systemic R2* fluctuations was 69.5% and 73.3% for SW1222 and LS174T, respectively (p=0.66, Mann-Whitney U test). RS(SD) ICA systemic z-scores (a) ICA tumour-specific z-scores 5 (c) z-score z-score 2.2 2.2 (e) mean ICA systemic timecourse mean ICA tumour specific timecourse a.u. a.u. (d) (b) 5 0 15 30 Time (minutes) 45 60 0 15 30 45 60 Time (minutes) Fig. 1 – SW1222 tumour xenograft. (a) Resting State standard deviation – RS(SD) map of R2* fluctuations. (b, c) ICA maps of systemic and tumour specific z-scores. (d, e) Mean timecourse curves of ICA decomposition for both systemic and tumour specific independent components. Fig. 2 – Percentage of tumour-specific and systemic voxels for each tumour type. Discussion & Conclusion: In this study, we have presented a novel method for separating systemic and tumour-specific tumour R2* fluctuations, based on independent component analysis (ICA) and pulse oximetry. The advantage of this approach over methods that directly correlate R 2* timecourses, and segment voxels into purely systemic or tumour-specific effects, is the ability of ICA to assess the relative contribution of systemic and tumour-specific effects in individual voxels. LS174T tumours revealed a greater proportion of voxels with tumour-specific and systemic fluctuations than SW1222 tumours, although these differences were not statistically significant. The relatively high percentage of voxels in both tumour-specific and systemic groups is possibly due to the ability ICA has to separate between the two effects. Clear differences have previously been observed in the hypoxia and vascular perfusion status of the tumours used in this study, with SW1222 tumours much better perfused and less hypoxic than LS174T tumours [7]. Further work is required to evaluate the details of mechanisms underpinning spontaneous, tumour specific fluctuations in each tumour type, which we anticipate will be elucidated with the novel methodology presented here. Acknowledgements: This work was carried out as part of King’s College London and UCL Comprehensive Cancer Imaging Centre CR-UK & EPSRC, in association with the MRC, DoH and British Heart Foundation (England). References: [1] Baudelet C, et al., Phys. Med. Biol. 49: 3389-411. [2] Brown JM, Br. J. Radiol. 52: 650-6. [3] Patan S, et al., Microvasc. Res. 51:260-72. [4] Mollica F, et al., Microvasc. Res. 65:56-60. [5] Hyvärinen A, IEEE Trans. Neural Netw. 10:626-34. [6] van de Ven, et al., Hum. Brain Mapp. 22: 165-78. [7] Folarin A, et al., Microvascular Res. 80: 89-98. ND-‐Track: Tractography utilising parametric models of white matter fibre dispersion Matthew C Rowe, Hui Zhang, Daniel C Alexander Centre for Medical Image Computing, Dept. Computer Science, University College London Introduction: This work develops a tractography algorithm to leverage fiber dispersion estimates derived from fitting parametric models of orientation dispersion to diffusion data. Tractography techniques are powerful tools to probe white matter (WM) connectivity non-‐ invasively. Most current techniques follow a small number of discrete directions per voxel to identify WM connections. This approach addresses the limitation of traditional DTI-‐based tractography for regions with crossing fibers. However, it remains an oversimplification for regions with fanning and bending configurations, where the underlying fiber orientation distributions are continuous rather than discrete [2]. Following only a discrete set of directions in this case misrepresents the underlying anatomy and is likely to result in false negative connectivity estimates. Recent parameterized models of fiber dispersion represent Figure 1: Synthetic phantom exhibiting dispersion (blue lines), tracking results (red lines) such sub-‐voxel fiber architecture more realistically and provide more accurate estimates of dispersion than non-‐parametric techniques such as spherical deconvolution, which are vulnerable to noise [3]. Here, we present a new tractography algorithm, hereby referred to as ND-‐Track (Neurite Dispersion Tracking), that leverages directional information gathered from parametric models of dispersion. We investigate the advantages of tracking with dispersion measures on a simple phantom and in in-‐vivo data, tracking through the coronal radiata, a region known to exhibit a significant degree of fiber dispersion. We further demonstrate that this approach does not compromise the tracking of the WM pathways for which the standard technique works well. Methods: We aim to sample directions from the dispersion patterns in each voxel while maintaining smoothness to produce biologically plausible tracks. Starting from a seed point, ND-‐ Track propagates a streamline by sampling permissible directions in each voxel it passes through. Orientation distribution functions (ODF) modelling dispersion are derived via the NODDI technique [1]. NODDI yields a dispersion pattern in each voxel in the form of a Bingham distribution. For a streamline entering a voxel with direction , a set of 500 candidate directions are sampled from the local Bingham distribution via a rejection sampling algorithm. The candidate directions are then weighted by a Watson prior distribution, with mean direction . This favours candidate directions that deviate minimally from the previous step while adhering to the locally permissible ND-‐Track CSD set of fibre orientations. One of these candidate directions is then selected probabilistically as the Figure 2: Tracking results from seed regions in next propagation direction and the track propagated in this direction by a step size of 1mm, or half the CC and the IC, a voxel. This ensures the propagation of smooth streamlines that follow directions sampled directly from dispersion-‐modelling distributions. Tracks are terminated upon entry into a grey matter mask, derived from a Freesurfer parcelation. All in-‐vivo tracking results are shown as maximum intensity projections thresholded at 1% of maximum intensity. Data: The diffusion data was acquired from a healthy male subject on a clinical 3T Philips system with isotropic voxels of 2mm. It consists of one 30 direction shell with b-‐value of 1000 s/mm^2, a 60 direction shell with b-‐ value 2000 s/mm^2, and 9 b=0 images with SNR about 20. Results and Discussion: Figure 1 a) shows how the algorithm performs on a numerical phantom exhibiting dispersion, Figure 1 b) shows the result of standard PICo tracking for comparison [4]. The blue lines outline the synthetic fiber structure of the phantom and the red lines show the tracking result from a single seed at the base of the image. The standard PICo tracking algorithm performs poorly on this dispersing phantom, only covering a very narrow range of connections. Utilising models of dispersion, we recover much more of the potential connections of the phantom. Figure 2 shows a comparison of results of ND-‐Track with tracking results Figure 3: ND-‐Track results on major white matter structures from constrained spherical deconvolution (CSD) using MRTrix from seed regions in the midbody of the corpus callosum (CC) and the internal capsule (IC). The techniques give comparable results, however two notable differences can be identified. CSD tracking more heavily favours the characteristic vertical connections from the CC, as indicated by the red arrows in Figures 2a and 2b. Furthermore, ND-‐Track gives a more even representation of connections spread throughout the cortex, this is most clear in Figures 2c and 2d, the white arrows indicate a region where lateral cortical connections are covered better by ND-‐Track. Figure 3 shows the major white matter pathways, for which the standard tracking techniques work well, successfully tracked by the algorithm. The images show successful recovery of the inferior longitudinal fasciculus (ILF), the superior longitudinal fasciculus (SLF), the cingulum (Ci) and the occipito-‐frontal fasciculus (OF). Conclusion: We present a tractography algorithm leveraging distributions modelling dispersion to propagate streamlines. ND-‐Track shows expected performance on major white matter structures and advantages in tracking through regions of fanning fiber structure, both in simple numerical phantoms and in vivo data. At this preliminary stage the algorithm does not consider crossing fiber configurations, however despite this limitation this preliminary investigation shows that results can be obtained comparable to state of the art multi-‐fiber techniques. In future work, a straightforward extension is planned to include multiple dispersing fiber populations per voxel. References: [1] Zhang et al. NIMG12 [2] House & Panksy 1960 [3] Sotoripoulos et al. NIMG 12 [4] Parker et al. JMRI03 Abstracts: Oral Presentations Brain Investigations 25 Optimisation of calibrated FMRI for the detection of regional alterations in absolute CMRO2 Alan Stone, Kevin Murphy, Ashley D Harris, Richard G Wise CUBRIC, School of Psychology, Cardiff University Background & Aims: Calibrated FMRI techniques providing absolute CMRO2 quantification in the brain have emerged recently [1, 2]. Using a BOLD signal model, similar to [3], simultaneous CBF and BOLD measures performed during hypercapnia and hyperoxia (HC-HO) enable estimates of Sv O2 and in turn absolute CMRO2 to be made allowing improved interpretation of metabolism changes in disease or drug states of the brain [4]. The aim of this study is to investigate the feasibility of a novel simultaneous respiratory manipulation for the evaluation of stimulus-induced changes in absolute CMRO2 and to compare these results with those made using a more standard respiratory manipulation. The standard respiratory manipulation was performed using interleaved HC-HO (intHC-HO), Fig. 1(A), whereas the novel respiratory manipulation involves simultaneously varying the HC-HO (simHC-HO) delivered to the subject, Fig. 1(B). By simultaneously varying HC-HO we can measure cerebrovascular properties, α and β which must be assumed when using intHC-HO and potentially increase sensitivity to detecting changes in baseline CMRO2 . B) Methods: 8 normal healthy participants (aged 24-39; mean A) age 33.5±5; 2 female) were scanned twice with a protocol lasting 50 mins on a 3T GE HDx MRI. In the first session absolute CMRO2 was estimated using intHC-HO and in the second session simHC-HO was used. For both HC-HO manipulations simultaneous CBF and BOLD image time-series were acquired using a dual-echo PASL PICORE QUIPSS II with a dual-echo gradient echo (GRE) readout and spiral kspace acquisition. A functional localiser lasting 5 mins 30 s was used to create ROI’s of activated voxels in visual and motor cortex (VC and MC). The respiratory challenges were Figure 1: HC-HO gas challenges with expected CBF and BOLD administered using dynamic end-tidal forcing[4] allowing in- time-series for (A), interleaved and (B), simultaneous designs. dependent control of partial pressures PET O2 and PET CO2 . Two CMRO2 acquisitions were performed, one with continuous visual and motor stimulation (task) and one at rest with the participant fixating on a cross hair overlaid on a black background (rest). Structural ROI’s of global grey matter (GM) and functional ROI’s in VC and MC were generated. From this estimates of M, Sv O2 and CMRO2 were produced using a Bayesian parameter estimation routine and BOLD signal model. 1$min$ 1$min$ +$0$mmHg$ 1$min$ 1$min$ +$200$mmHg$ +"200"mmHg" 2"min" PETO2$ 2"min" 2"min" PETO2" +$7$mmHg$ +7"mmHg" 2.5$mins$ 2"min" +4"mmHg" 2.5$mins$ 2"min" PETCO2$ +$1$mmHg$ 2"min" +"0"mmHg" +"1"mmHg" +2mmHg" +8"mmHg" +6"mmHg" 2"min" 2"min" +3"mmHg" 2"min" 2"min" PETCO2" CBF"Signal" CBF$Signal$ BOLD$Signal$ BOLD"Signal" 18$min$ 0$min$ 18"min" 0"min" Results: (A) (B) ROI GM intHC-HO fVC fMC GM simHC-HO fVC fMC Figure 2: CMRO2 maps of a single subject using intHC-HO, (A) and simHC-HO (B). Colorbar ranges from 0-250 µmol/100g/min CBF ml/100g/min Sv O2 CMRO2 µmol/100g/min -Rest -Task -Rest -Task -Rest -Task 51.8±7.7 53.6±5.3 62.3±11.9 84.0±18.5* 46.7±18.8 62.3±9.5* 0.55±0.17 0.61±0.12 0.51±0.07 0.60±0.18 0.58±0.13 0.70±0.17 187.5±79.4 164.6±50.7 243.4±57.9 253.5±83.6 140.8±54.0 140.6±74.0 -Rest -Task -Rest -Task -Rest -Task 49.6±10.2 52.4±11.0 57.2±13.1 79.3±21.0 * 39.7±14.3 59.0±18.1 * 0.64±0.11 0.67±0.11 0.65±0.11 0.65±0.08 0.66±0.13 0.68±0.08 137.7±38.0 139.8±55.9 152.0±26.0 213.9±43.2* 104.4±45.5 148.9±61.3* Table 1: Group average CBF, Sv O2 and CMRO2 estimated using intHC-HO and simHC-HO. * indicates significant difference, p<0.05, between rest and task conditions. Discussion: Whole brain absolute CMRO2 maps were produced using both HC-HO manipulations, Fig. 2. From Table 1, simHC-HO group average GM CMRO2 values compared more favourably to established literature values of 146±30 [1] and 145±30 [2] µmol/100g/min. SimHC-HO was also capable of detecting significant (p<0.05) increases from baseline CMRO2 with stimulation in VC(40%) and MC(43%). In contrast no significant changes in CMRO2 were detected in stimulated areas using intHC-HO. Conclusion: By performing an optimised absolute CMRO2 routine using a simultaneous HC-HO manipulation we are able to fit for parameters, α and β using a BOLD signal model allowing significant differences in baseline CMRO2 to be detected in stimulated regions. References:1. Bulte et al . NeuroImage. 2012; 60: 582; 2. Gauthier et al. NeuroImage. 2012; 60:1212; 3. Davis et al. PNAS. 1998, 95:1834; 4. Wise et al. JCBFM. 2007, 27:1521-1532; Viable and Fixed White Matter: DTI and Microstructural Comparisons at Physiological Temperature. S. Richardson1, 2, B. Siow1, 2, E. Panagiotaki2, T. Schneider3, M.F. Lythgoe1, D.C. Alexander2 1 UCL Centre for Advanced Biomedical Imaging, Division of Medicine and Institute of Child Health, University College London, United Kingdom, 2 Centre for Medical Image Computing, Dept of Computer Science, University College London, London, United Kingdom. 3 Institute of Neurology, University College London, London. Target Audience: Researchers in diffusion weighted MRI. Specifically: microstructural modelling, tractography, and sequence development. Abstract: We compare fixed and viable isolated tissue (VIT) in identical conditions at physiological temperature. We acquired DTI data sets with a range of acquisition parameters and a rich multi-b-value diffusion weighted MR (DW-MR) dataset for microstructural tissue model fitting. DTI data demonstrated a significant increase in radial diffusivity (RD) in fixed samples in comparison to VIT. Model fitting demonstrated that similar models best explain data from both samples and differences in parameter estimates reflect DTI measured RD and MD differences. The stability of the model ranking allows us to conclude that fixed tissue is a reasonable model for in-vivo, although significant differences in the fitted model parameters (and DTI measured FA and RD) suggest that water in individual compartments within the tissue behaves quite differently. Purpose\Background: Fixed tissue samples are used for testing and validation of DW-MR methods e.g. (1,2). Fixation affects measured diffusivity in neuronal tissue (3,4). VIT mitigates several confounders of in vivo MR acquisitions e.g. movement, vascular and susceptibility effects (5). Comparing VIT with fixed tissue in identical conditions provides information on the efficacy of fixed tissue as a model for viable tissue. In this work we used a VIT maintenance system (6) (Figure 1) to investigate DW-MR detectable differences between VIT and fixed tissue at physiological temperature. DW-MR is highly sensitive to temperature differences (7), previous comparisons between viable and fixed tissue have been conducted at lower temperatures than in this work (3), where we use physiologically realistic temperatures. We acquired: (A) various DTI datasets designed to probe the dependence of DTI indices on acquisition parameters, both within and between VIT and fixed tissue; (B) a rich multi-b-value DW-MR dataset with which we fit a large range of multi-compartment tissue models to see which best explain data from both fixed and VIT samples (see (2) for full model list). Figure 1: MRI compatible incubation Methods: VIT: rat optic nerves (♂ Sprague Dawley) were maintained at 36.5°C and perfused with chamber (exploded). Diameter 26mm. a) inserts & bench support the optic oxygenated artificial cerebral spinal fluid (aCSF). Fixed nerves were prepared by immersion in 4% nerve, b) inflow, c) preheating system formaldehyde solution (24 h) then washed in phosphate buffered saline (10 h). (A) Diffusion tensor imaging in lower section, d) outflow and e) (DTI) experiments (n = 2 & 2): 30 direction pulsed-gradient spin-echo (PGSE) experiments, diffusion times isocenter positioning bar. (Δ): 10, 20 & 30ms, gradient durations (δ): 4, 6 & 8ms and gradient strengths (G): 32 – 71 G cm-2. [Exp 1: fix Δ & δ, alter G | Exp 2: fix G, alter Δ & δ | Exp 3: fix G, maintain b-value, alter Δ] TR: 1.5s, TE: minimised. (B) Multi-b-value DW-MR dataset from VIT and fixed nerves (n = 1& 1): [Δ: 10 - 50ms (sets of 10ms) | δ: 1.5 & 3ms | G: 4 – 40 G cm-2], 3 perpendicular & 1 parallel direction, TR: 2.0s, TE: minimised. All MR experiments were performed with a 9.4T Agilent VNMRS system. Scan durations were: 6 h (A) and 7h (B), matrix: 48 x 48, FOV: 6 x 6 x 2mm [x, y & z]. The open source Camino diffusion MRI tool kit (8) was used to fit DTI and the set of multi-compartment signal models from (2) to the DW-MR data. Results: Table 1 shows DTI calculated FA RD (µm2/s) Table 1: average FA and RD in VIT and Fixed Average Fixed and VIT DTI indices VIT Trend Fixed Trend VIT Trend Fixed Trend samples. RD in fixed samples was found to be significantly higher than that of VIT for all Exp 1: fix Δ & δ, increase G 0.78 + ve 0.70 + ve 242 - ve 279 -ve measurements (P<0.01, t-test). In both VIT Exp 2: fix G, increase Δ & δ 0.80 + ve 0.69 + ve 237 - ve 279 - ve and Fixed samples, Bayesian information Exp 3: fix G, increase Δ, decrease δ 0.79 none 0.69 none 236 none 294 none criterion (BIC) ranking of fitted models was similar and selected three compartment models as the best performing (data not shown). In the terminology of (2), tensor and zeppelin models outranked balls (hindered compartment) while stick models outranked cylinder models (restricted compartment). FA and RD from the best fitting compartment models coincide with DTI detected differences between fixed and VIT samples. Volume fractions estimated by the best fitting models were comparable. Discussion: A temperature difference of 17ºC has been shown to reduce extracellular ADC by 25% (9), comparing samples at physiological temperature provides more accurate information on differences than previous work e.g. (4). These clear variations and differences in DW-MR parameters should be taken into account when comparing tissue states and studies using different acquisition parameters. Differences in extracellular ADC between VIT and fixed tissue have been previously demonstrated in brain slice preparations at room temperature (3), however this is the first demonstration of increased in RD in fixed white matter in comparison to VIT in identical conditions at physiological temperature. Fixation produces similar model rankings but alters their parameters, suggesting that the broad tissue structure is maintained but individual water populations behave differently. Conclusion: When measured in identical conditions, fixed tissues DW-MR properties depart significantly from those of viable tissue; however similarity of model rankings suggests that DW-MR detectable tissue features are maintained. We conclude that while VIT has clear advantages (see (6) for discussion), fixed tissue is suitable, when used cautiously, as a basic test-bed for DW-MR development. Further work will increase DTI repeats to elucidate acquisition parameter dependant patterns in experiments 1-3 in both VIT and fixed tissue. Selected References:(1) Assaf Y, Blumenfeld-Katzir T, Yovel Y, Basser PJ. Magn. Reson. Med. 2008;59:1347–1354. doi: 10.1002/mrm.21577.(2) Panagiotaki E, Schneider T, Siow B, Hall MG, Lythgoe MF, Alexander DC. NeuroImage 2012;59:2241–2254. doi: 10.1016/j.neuroimage.2011.09.081.(3) Shepherd TM, Thelwall PE, Stanisz GJ, Blackband SJ. Magn. Reson. Med. 2009;62:26–34. doi: 10.1002/mrm.21977.(4) Sun S-W, Liang H-F, Le TQ, Armstrong RC, Cross AH, Song S-K. NeuroImage 2006;32:1195–1204. doi: 10.1016/j.neuroimage.2006.04.212.(5) Smith TB, Nayak KS. Imaging Med 2010;2:445–457. doi: 10.2217/iim.10.33.(6) Richardson S, Siow B, Batchelor AM, Lythgoe MF, Alexander DC. Magn Reson Med 2012:n/a–n/a. doi: 10.1002/mrm.24410.(7) Harris KR, Woolf LA. J. Chem. Soc., Faraday Trans. 1 1980;76:377–385.(8) Cook P, Bai Y, Gilani N, Seunarine K, Hall M, Parker G, Alexander D. Camino: OpenSource Diffusion-MRI Reconstruction and Processing. In: 14th Scientific Meeting of the International Society for Magnetic Resonance in Medicine. ; 2006. p. 2759.(9) Thelwall PE, Shepherd TM, Stanisz GJ, Blackband SJ. Magn. Reson. Med. 2006;56:282–289. Measuring Tissue Perfusion in the Human Brainstem in Normo- and Hypercapnia Using Multi Inversion Time Pulsed Arterial Spin Labelling 1 1 1 2 3 1 Esther A.H. Warnert, Ashley D. Harris , Kevin Murphy , Michael Chappell , Judith E. Hall , Richard G. Wise 1 Cardiff University Brain Research Imaging Centre, Cardiff University, Cardiff, The United Kingdom, 2Institute of Biomedical Engineering, University of Oxford, Oxford, The United Kingdom, 3School of Medicine, Department of Anaesthetics, Intensive Care, and Pain Medicine, Cardiff University, Cardiff, The United Kingdom Introduction The brainstem is involved in control of critical physiological processes, including cardiovascular function and breathing. Currently, arterial spin labelling studies often exclude the brainstem due to difficulties with the acquisition and analysis of perfusion signal in this brain region. Here, we aim to demonstrate the measurement of cerebral blood flow (CBF) in the brainstem during normocapnia by separating the macrovascular and tissue perfusion signal using a two-compartment model containing both arterial and tissue components [1]. Multi-inversion time pulsed ASL (MTI pASL) with short echo-time spiral read out was used to minimize signal drop out in the brainstem. Preliminary results of validating our MTI pASL method by measuring brainstem perfusion during a hypercapnic challenge, which is expected to increase CBF in the whole brain, are included as well. Methods Seven young, healthy subjects (3 female, mean age 29.6 ± 4.9 years) were recruited. Normocapnia perfusion measurements were made using the PICORE tagging scheme [2] with an echo-planar spiral read-out with 2 interleaves per image. Ten control-tag pairs were acquired for each of 13 inversion times (TI): 0.1s, 0.2s, 0.3s, 0.4s, 0.5s, 0.6s, 0.7s, 1.0s, 1.3s, 1.6s, 1.9s, 2.2s 2.5s. A QUIPSSII [3] cut-off of the magnetic label at 0.7s was used for TI > 0.7s. Other image parameters were: label thickness 20 cm, 1 cm gap between label and most proximal imaging Figure 1. Examples of kinetic curves fitted by the model. Red: arterial 3 slice, resolution 3x3x7 mm with a 1mm slice gap, 13 slices, compartment, blue: tissue compartment, black: total model. Green: Δtart and Δttiss. echo time 3.2ms. Repetition time (TR) was minimised for each individual TI. For quantification purposes, a M0 image was acquired with the same imaging parameters, except an 3 infinite TR and no labelling pulse. For each TI a difference 2 1 image was obtained by subtracting the 10 tag-control pairs and averaging the subtractions. Kinetic curves of the 0 85 1 2 3 magnetic label were fitted to the average TI images with Figure 2. Example of normocapnia CBF maps (ml/100 gr/min). Left: Sagittal slice, methods described by Chappell et al [1], using BASIL within including outline of used brainstem mask in green. On the right: Axial slices 1, 2, and the FSL toolbox v5.0 (http://fsl.fmrib.ox.ac.uk). In voxels 3, corresponding to the red lines in the sagittal slice. that contained arterial signal, only the tissue perfusion Table 1. Average parameters resulting from fitting a two-compartment model of curve was considered (Figure 1). Individual brainstem and perfusion to MTI pASL data, acquired in normocapnia. aBV Δtart Δttiss CBF whole brain gray matter masks were used to analyse kinetic ROI (%) (ms) (ms) (ml/100g/min) curves, perfusion (CBF), tissue and arterial arrival times Brainstem 0.68 ± 0.54 438.0 ± 20.4 701.8 ± 55.6 32.1 ± 4.0* (Δttiss, Δtart), and arterial blood volume fraction (aBV). Gray matter 0.40 ± 0.10 443.6 ± 7.3 701.5 ± 36.9 53.4 ± 8.4 Results and Discussion Figure 2 shows an example of CBF maps in normocapnia. Average brainstem CBF is significantly *Significant difference between gray matter and brainstem (paired t-test, p<0.001) lower (paired t-test, p<0.001) than gray matter CBF (Table 1), which is 100 Normocapnia Hypercapnia expected as the brainstem contains both gray and (less perfused) white matter. As expected based on the local vascular anatomy, the average macrovascular arterial signal contribution per voxel is larger in the brainstem than in the gray matter (aBV, Table 1), although this is not significant. Higher values for Δtart (Table 1) were found than previously reported values ( ̴300 ms) [4, 5], possibly due to the use of 0 a two-compartment model that accounts for a rapidly increasing arterial component, instead of using a single compartment model that Figure 3. Example of CBF maps for one subject in normo- and only contains a slower increasing tissue perfusion component and hypercapnia. CBF in ml/100 gr/min. Outline of brainstem mask in green. thus leads to shorter arterial arrival times. Preliminary Hypercapnia: Methods and Results Three healthy males (age 29.0 ± 5.6 years) are recruited so far for the hypercapnia experiment. CBF maps are obtained while the subject is breathing medical air (normocapnia) or a mixture of 5% CO2 and air (hypercapnia). To shorten the time spent in hypercapnia, only 8 control-tag pairs were acquired for 6 TIs (0.2s, 0.4s, 0.6s, 1.0s, 1.6s, 2.2s). The other acquisition parameters and the analysis remained the same as described above. Examples of CBF maps for one subject are shown in Figure 3. Average cerebral vascular response (% CBF increase per increase in end-tidal CO2 pressure) is 7.4 ± 3.8 % / mmHg in whole brain gray matter and 15.5 ± 14.7 % / mmHg in the brainstem. Summary The current study has shown that using a two-compartment perfusion model with short-echo time spiral read out pASL enables kinetic curve fitting and provides an estimate of brainstem CBF in normocapnia by separating the tissue signal from the macrovascular arterial signal. Preliminary results of the validation of the MTI pASL method include estimates of cerebral vascular response in the brainstem. References 1.Chappell, MA, et al., Magn Reson Med, 2010. 63. 1357-65. 2.Wong, EC, et al., NMR Biomed, 1997. 10. 237-49. 3.Wong, EC, et al., Magn Reson Med, 1998. 39. 702-8. 4.Chen, Y, et al., MAGMA, 2012. 25. 135-44. 5.Huang, AJ, et al. Proceedings of ISMRM. 2011, 19, 301. Probing Spinal Cord Microstructure with the Neurite Orientation Dispersion and Density Imaging Francesco Grussu1,3, Torben Schneider1,3, Hugh Kearney1, Hui Zhang3, David H. Miller1, Olga Ciccarelli2, Daniel C. Alexander3 and Claudia Angela M. Wheeler-Kingshott1 1 NMR Research Unit, Department of Neuroinflammation, UCL Institute of Neurology, University College London 2 NMR Research Unit, Department of Brain Repair and Rehabilitation, UCL Institute of Neurology, University College London 3 UCL Department of Computer Science and Centre for Medical Image Computing, University College London Background and aims Diffusion tensor imaging (DTI) provides a number of indices sensitive to tissue microstructure, which however only yield summary information as they cannot disentangle the contribution of specific features such as coherence and density of axon bundles [1]. As these features may be important biomarkers for diseases like Multiple Sclerosis (MS) [2], neurite orientation dispersion and density imaging (NODDI) is a promising new technique, being capable of providing estimates of such quantities [1]. However, its performance in the spinal cord is still unknown. Here, we firstly demonstrate the feasibility of NODDI in the spinal cord, comparing the technique to DTI on single shell diffusion-weighted (DW) MRI data of healthy volunteers (HVs) [3]. Then, a pilot scan implementing the full NODDI protocol [1] is performed on a HV and analysed. Figure 1: a slice of some maps estimated in a HV with one b shell. a) and b) from DTI, c) and d) from NODDI. MD is in µm2 ms–1. Figure 2: in a), ODI histogram from all HVs (single shell data) with separation of higher (green) and lower (blue) values. In b), overall HVs’ normalized FA histograms evaluated for highly (green) and less (blue) dispersed tissue. In c), a HV’s ODI slice, thresholded in d) with the Otsu’s method (threshold 0.088). Methods Single shell data DW MRI scans of the cervical spine of 10 HVs (4 m, mean age 34, SD 9.10) were studied [3]. The imaging protocol consisted of 30 b=1000 s mm–2 DW volumes and 3 b=0 volumes. On such data, NODDI and DT models were fitted using a freely available NODDI Matlab Toolbox for the former and Camino [4]. NODDI fractional isotropic volume (FisoV) was fixed to 0 to account for the reduced number of shells [1]. For DTI, fractional anisotropy (FA), mean, axial and radial diffusivities (MD, AD, RD) were evaluated; for NODDI, fractional intracellular volume (FicV) and orientation dispersion index (ODI). The indices were then analysed as follows: a) Pearson’s correlation coefficients between DTI and NODDI indices were calculated; b) the histograms of the indices were calculated, and the Otsu’s method [5] was applied on the ODI one after removing the few occurrences of ODI=1, likely due to partial volume effects on the cord boundary. Double shell data A pilot DW MRI scan of a female HV (age 32) was Figure 3: 2 shell pilot scan. In a), the second slice in the first b=0 volume. In b), the same slice in the sixth b=0 volume. In c), b) registered back to a). performed at cervical level, following the published two-shell NODDI protocol [1]. The implementation in FSL of a motion correction stage was necessary due to the longer acquisition time. It consisted of a slice-wise linear registration, known to be effective in the spinal cord [6]. On the registered data, both DT and NODDI models were fitted, using the b = 711 s mm–2 shell for the former. The derived metrics were then visually inspected. Results Single shell data DTI and NODDI maps showed a core resembling grey matter (figure 1), which was best enhanced in ODI. FicV exhibited a good correlation with all DTI measures (strongest with RD, r = – 0.874; weakest with AD, r = – 0.650), whereas a weaker correlation was seen for ODI (highest with RD, r = 0.453) and between FicV and ODI (r = – 0.377). The histogram analysis proved that the distributions of DTI maps and FicV exhibited a single peak, whereas a second peak was noticeable in ODI (figure 2.a). Such findings suggest the existence of two potential tissue fractions, i.e. high and less dispersed. The overall HVs’ ODI histogram was then thresholded with the Otsu’s method (threshold 0.088), which provided two clusters resembling grey and white matter (figure 2.c and 2.d). However, the FA distributions of the two groups of voxels were similar and partly overlapping (figure 2.b). Double shell data The motion correction stage provided robust results on visual inspection (figure 3). Furthermore, we could observe a core resembling grey matter in both DTI and NODDI maps (figure 4); such core was well delineated in ODI (figure 4.d), proving its sensitivity to local tissue coherence. Lastly, we observed that a certain amount of isotropic signal was needed to explain the NODDI model, as FisoV did not vanish completely within the cord (figure 4.c). Conclusions In this work we demonstrate the feasibility of NODDI in the spinal cord. In the single shell data, the analysis identified FicV as the primary source of variation contributing to the patterns of DTI indices. In contrast, ODI enhanced the differences between two clusters resembling grey and white matter. The two-shell data instead provided meaningful NODDI maps, despite the longer acquisition time which required correction for motion artifacts. In conclusion, NODDI did not just replicate standard DTI measures but provided additional information, even with just one b shell [3]. Future work is required to quantify the effect of the reduction of the number of shells and to obtain an overall exhaustive validation of NODDI in the spinal cord. Figure 4: derived maps from the 2 shell pilot scan. In a), a slice of DTI FA. From b) to d), NODDI maps of the same slice. Reference [1] Zhang H et al. NeuroImage (2012); 61(5):1000–16 [2] Schmierer K et al. NeuroImage (2007); 35(2): 467–77 [3] Grussu F et al. Proc of ISMRM (2013, in press) [4] Cook PA et al. Proc of ISMRM (2006) [5] Otsu N. IEEE Trans Syst, Man, Cybern (1979); 9(1):62–66 [6] Mohammadi S et al. NeuroImage (2013); 70: 377-85. Acknowledgment This work was funded by the MS Society for Great Britain and Northern Ireland and the UCL Grand Challenge Studentships scheme and supported by researchers at the National Institute for Health Research University College London Hospitals Biomedical Research Centre. Clinical utility of NODDI in assessing patients with epilepsy due to focal cortical dysplasia Gavin P Winston1, Mark R Symms1, Daniel C Alexander2, John S Duncan1, and Hui Zhang2 1 Epilepsy Society MRI Unit & Department of Clinical and Experimental Epilepsy, UCL Institute of Neurology, London 2 Department of Computer Science & Centre for Medical Image Computing, University College London Background/Aims: DTI indices such as FA reflect many underlying parameters including neuronal density, fibre orientation dispersion, axonal diameter and degree of myelination. DTI assumes Gaussian diffusion within a single compartment and does not adequately reflect the microstructure within a voxel. Multi-compartment models more accurately reflect the diffusion MR signal by distinguishing restricted non-Gaussian diffusion (intracellular) and hindered Gaussian diffusion (extracellular space)1 but typically require scans times far longer than clinical feasible or make invalid assumptions (e.g. no axonal dispersion). The NODDI (neurite orientation dispersion and density imaging) model distinguishes two key variables contributing to FA changes - neurite density and orientation dispersion - with a clinically feasible scan protocol2. We apply NODDI in a clinical population of patients with epilepsy due to focal cortical dysplasia (FCD) to show that additional parameter estimates are compatible with underlying disrupted tissue microstructure and provide useful additional clinical information. Methods: Patients with FCD were scanned on a 3T GE Signa HDx scanner with a full clinical protocol and a NODDI protocol optimized for the scanner3 (single shot EPI, 50x2.5mm axial slices, TE 85ms, TR 13s, 9 b=0 s.mm-2, 24 b=700 s.mm-2, 48 b=2000 s.mm-2, maximum gradient strength 40mTm-1, slew rate 150Tm-1s-1, scan time 20 mins). The NODDI Matlab Toolbox4 was used to fit a three-compartment model (intracellular, extracellular, CSF) yielding estimates of the intracellular volume fraction (ICVF). Results: In patients with FCD, intracellular volume fraction was reduced (Figure 1) compatible with iontophoretic studies of resected human tissue5. The same was true in tuberous sclerosis which has the same pathology as FCD (Figure 2). The affected area was easier to identify than on corresponding fractional anisotropy (FA) or mean diffusivity (MD) images and was clearly seen even when it was hard to identify on anatomical images (Figure 3). Figure 1: FCD with previous partial resection: T1weighted image (left); reduced ICVF (right) Figure 2: Tuberous sclerosis (tubers have the same pathology as FCD) - T1-weighted image (left) with corresponding reduced ICVF (middle, sagittal and right, coronal) Figure 3: FCD poorly defined on anatomical images including T1-weighted (far left), T2-weighted PROPELLER (second from left) and hard to discern on standard DTI images including fractional anisotropy (FA, middle), mean diffusivity (MD, second from left) but easily visible as reduced ICVF (far right) Discussion: In patients with epilepsy, identifying the location of the epileptogenic zone is critical in planning surgical treatment but up to 20-30% have normal MRI scans. Many patients have undetected FCD, a developmental anomaly characterized by disrupted laminar architecture/columnar organisation and abnormal cells7. The classical neuroimaging findings described on T1and T2-weighted images are not always present8. DTI changes including reduced FA and increased MD in underlying white matter are non-specific, extend beyond the area of abnormality9 and cannot evaluate dysplastic grey matter due to the low FA and CSF contamination. The NODDI model is suitable for both grey and white matter and by modeling CSF as a separate compartment avoids CSF contamination. A key fitted parameter is the intracellular volume fraction. Iontophoretic studies have shown that the extracellular volume fraction is increased (and thus intracellular volume fraction in reduced) in human neocortical tissue removed during surgery in patients with FCD5. Findings consistent with this have been demonstrated on NODDI scans and can be more clearly demonstrated than on other clinical or diffusion sequences. Conclusion: NODDI is viable to apply to a clinical population and the findings of reduced intracellular volume fraction are compatible with the known pathology of FCD. NODDI may assist in patients with epilepsy the clinical identification of areas of FCD not seen on other imaging sequences. References: 1Panagiotaki et al (2012) Neuroimage 59(3):2241-54, 2Zhang et al (2012) Neuroimage 61:1000-16, 3Alexander (2008) Magn Reson Med 60(2):439-48, 4http://cmic.cs.ucl.ac.uk/mig/index.php?n=Tutorial.NODDImatlab, 5Vargova et al (2011) Neurosci Lett 499(1):19-23, 7Blumcke (2009) Epileptic Disord 11:181-93, 8Barkovich et al (1996) J Clin Neurophys 13(6):48194, 9Eriksson (2001) Brain 124(3):617-26. Environmental Enrichment retards the development of neuropathology and ameliorates motor deficits in a mouse model of Huntington’s Disease: an in vivo MRI study Jessica Steventon1,2, David Harrison2, Anne Rosser2, Rebecca Trueman3, Simon Brooks2, Derek Jones1 1 CUBRIC, Cardiff University, 2 School of Biosciences, Cardiff University, 3 School of Biomedical Sciences, University of Nottingham. Introduction Huntington’s Disease (HD) is a fatal neurodegenerative disease, characterised by atrophy in the striatum, and the development of cognitive and motor deficits. Environmental enrichment has been found to delay disease progression in mouse models of HD1. The neurobiological basis underpinning this functional gain has yet to be studied. In vivo macrostructural (T2-weighted) MRI has previously been applied in mouse models of HD to study change in brain tissue volume. However, in vivo diffusion MRI - to assess tissue microstructure, and diffusion tractography - to reconstruct white matter pathways, have not yet been applied pre-clinically with algorithms capable of resolving crossing fibers. Our objective was to investigate whether environmental enrichment produced alterations in brain macrostructure and/or microstructure, detectable with in vivo T2-weighted MRI and diffusion MRI respectively. Materials and Methods YAC128 transgenic HD mice and age-matched wild-type (WT) mice were exposed to an enriched environment comprising of daily cognitive training on a serial implicit learning task2 and dietary restriction throughout their lifetime. At 21 months old, 22 trained mice (11 YAC 128, 11 WT) and 22 untrained mice (11 YAC128, 11 WT) underwent in vivo MRI on a 9.4 T magnet using a 2D Rapid Acquisition with Refocused Echoes (RARE) T2-weighted sequence: FOV = 15.4 x 15.4 mm, slice thickness = 400 µm, 30 slices, in-plane resolution = 120 x 120 µm, TR/TE=4000/35 ms. In a subset of animals (n = 32), a 4-shot DTIEPI sequence was also acquired: 27 slices of thickness = 320 µm, FOV = 22.4 mm x 22.4 mm, acquisition matrix = 96 x 96, TR/TE=14604/TE 20 ms, with diffusion weighting (δ=4ms, Δ=10ms, b = 1000 s/mm 2) in 30 directions and 5 non-diffusion weighted images. Following this, a battery of behavioural tests was performed on all animals (rotarod test, locomotor activity, watermaze). Regions of interest (ROI’s) were drawn manually for the striatum and cortex in 5 slices that demonstrated clear anatomical landmarks. Diffusion weighted images were corrected for motion/distortion and partial volume contamination 3. Tractography based on constrained spherical deconvolution4 (CSD, lmax = 6) was performed in order to resolve multiple intra-voxel fiber orientations and mean tensor-based parameters and tract volume were obtained for the cortico-striatal, callosal and anterior commissural pathways. Results Fig. 1. Training-induced macrostructural change. Volume of striatal ROI in WT and transgenic mice. Error bars represent 1 standard deviation (SD). Striatal volume (left and right) and cortical volume (not shown) was significantly lower in transgenic animals, p<0.001. Whole brain volume was not affected by genotype. A disease-modifying effect of enrichment was seen in transgenic mice; neuropathology in the striatum was reduced in both hemispheres, with and without correction for whole brain volume, all p<0.05. There was no main effect of enrichment in the cerebral cortex. . Fig. 2 (left). Training-induced microstructural change. Fractional anisotropy (FA) was significantly higher in enriched/trained transgenic mice compared to untrained transgenic mice, in the pathways between the striatum and motor cortex in the left hemisphere, p<0.05, reconstructed using CSD-based tractography. There was no significant difference in the other diffusivity indices or tract volume. There was no effect of training in the genu of the corpus callosum or anterior commissure. Error bars = 1 SD. Fig. 3 (middle and right). Training-induced motor recovery. Middle. An effect of genotype (F(1,38) = 6.61, p<0.05) and enrichment (F(1,38) = 9.29, . activity over 30 minutes; over 24 hours a genotype*enrichment p<0.01) is seen on motor interaction emerges, F (1,38)=4.20, p<0.05; not shown. Right. On the rotarod test, latency to fall (secs) was significantly improved at p<0.05, in mice exposed to enrichment compared to non-enriched mice, with a Genotype*Enrichment interaction, F (1,33)=5.40, p<0.05. There was no effect of enrichment on watermaze performance (not shown). Error bars = 1 SD. Discussion. Three converging outcome measures demonstrate a beneficial disease-modifying effect of environmental enrichment in a transgenic mouse model of HD: a reduction in neuropathology in the striatum, an increase in fractional anisotropy in striatal-cortical tracts, and functional gains in motor performance. This is the first study to successfully apply in vivo diffusion tractography in a mouse model of HD, and the first to demonstrate sensitivity of tract-specific measurements to detect microstructural changes following a behavioural intervention in a mouse model of disease, with implications for translational research and the evaluation of therapeutics. Further research is required to untangle the relative contribution of cognitive stimulation and dietary restriction. References 1Wood N et al. Neurobiol Dis. 2011; 42(3):427-37. 2Trueman R et al. Eur J Neurosc. 2007; 25(2):551-8. 3Pasternak O et al. MRM. 2009; 62(3):717-30. 4Tournier JD et al. Neuroimage. 2007; 1; 35:1459-72. Acknowledgments. Funded by the Wellcome Trust and EHDN. Imaging pH changes in piglet brain after acute hypoxia-ischemia using Amide Proton Transfer (APT) 1 1 2 2 3 3 3 1 3 3 3 M. Rega , F. Torrealdea , A. Bainbridge , D. Price , K. Broad , I. Fierens , M. Ezzati , A. Oliver-Taylor , R. Burnett , C. Uria , N. Robertson , S. 4 1 1 Walker-Samuel , D. Thomas , X. Golay 1 2 3 Institute of Neurology, UCL, London UK. Medical Physics and Bioengineering, UCLH, London UK. Institute of Woman’s Health, UCLH, 4 London UK. Centre of Advanced Biomedical Imaging, UCL, London UK. Background Hypoxia-ischemia (HI) in the new born infant is an important cause of death with 4 in 1000 neonates suffering asphyxiation before or at birth resulting in neonatal encephalopathy (NE) with a survival rate of 40% of which at least 25% suffer long term neurodevelopmental sequelae. Following HI, brain undergoes dramatic metabolic disturbances which lead to alterations in the pH of the tissues. Change in tissue pH is a good biomarker of the severeness of the condition; therefore it is important to develop techniques that track brain pH changes [1,2,3]. P-31 NMR spectroscopy, a well-established MR technique, allows quantitative pH measurements using the following form of the Henderson-Hasselbalch equation (Petroff et al., 1985a): (Equation 1), in which δ is the chemical shift between Phosphocreatine (PCr) and Inorganic Phosphate (Pi) peaks [4,5]. However, the main limitation of this technique is that it offers no spatial resolution. Amide proton transfer (APT) has the ability to indirectly detect protein concentrations in soluble form, through chemical exchange of the amide groups with the free water surrounding them. Transfer rate between amide protons and water is pH dependent and follows a basecatalyzed amide proton exchange relationship (Zhou et al., 2003): (Equation 2), where is the exchange rate of amide protons [6].Typically 1 unit of pH drop reduces the exchange rate by 90%, [7] making APT contrast very sensitive to the local pH. Aim Non-invasive mapping of regional pH changes in piglet brain undergoing neonatal Hypoxic-Ischemic insult using APT-MRI. Methods Piglets were surgically prepared [9] within 24 hours of birth and both common carotid arteries were isolated and encircled by remotely controlled vascular occluders. Piglets were then placed into the MRI scanner were anesthesia was maintained by the combination 31 of 2% isoflurane, nitrous oxide and continuous infusion of morphine (0.05mg/kg/hr) [1]. APT scans and global P MRS were acquired preand 60 minutes post- induced HI by occlusion of both common carotid arteries [8]. o APT protocol: The APT sequence consists of a saturation train of 80 Gaussian pulses (pulse length=50ms, FA=400 , 91% duty cycle), followed o 2 by a turbo-flash readout (TR=4.14ms, TE=2.09ms, FA=10 , FOV=100x100mm , matrix=128x128, slice thickness=4mm). Saturation was applied at ±6ppm for 77 frequency offsets. Reference scans were also acquired for normalization purposes. 31 P-31 NMR spectroscopy protocol: P MRS was acquired with an elliptical transmit-receive surface coil. The sequence consist of a global (hard) pulse with TR=10s. PH-weighted images: pixel by pixel Z-spectra were analysed at 3.5ppm from water (APT peak). The height of the peak was measured in order for pre and post insult maps to be calculated. The percentage change in APT was calculated by: (eq3). Results Fig1 shows the Z-spectra (pre- & post- insult) from a selected region in the brain. A clear disappearance of the amide peak is observed following insult compatible with a decrease in the exchange rate of Amides. As the exchange rate is base-catalyzed (equation 2), it represents a pH reduction in the corresponding region. Fig2 is a map of the percentage change in APT peaks of a piglet brain with induced HI, calculated by equation 3. There is a dramatic decrease in APT signal which corresponds to a decrease in the pH of the tissue, expected in ischemic episodes. 31 Furthermore P NMR spectroscopy (fig3) confirms the global acidification of the brain. (ΔpH=-1.7 calculated by the shift of inorganic phosphate peak with reference to phosphocreatine, equation 1). Conclusions Brain pH changes can be mapped using the APT technique described. The data from fig2, which demonstrate a 31 decrease in APT signal, are confirmed with P spectroscopy pH values (fig3). In conclusion, this study shows that the APT technique is suited for the mapping of pH changes in the brain of piglet undergoing Hypoxia-Ischemia. However for pH quantification further work is required to establish a reliable measure. References: [1] O.Iwata et al., Ann Neurol 58, 75(2005), [2] R.C.Vannucci, Pediatrics 85, 961(1990), [3] N.J.Robertson et al., Ann Neurol 52, 732(2002), [4] P.L.Hope et al., J.Neurochem 49, 1(1987), [5] S.R.Levine et al., Radiology 185, 537 (1992), [6] P.Z.Sun et a;., NeuroImage 60, 1(2012), [7] J.Zhou et al., Transl Stoke Res 3, 76(2011), [8] N.J.Robertson et al., Brain 136, 90(2013), [9]. A.Lorek et al., Pediatr Res 36, 699(1994). Novel Biomarkers of Tau pathology in a mouse model of Alzheimer’s Disease: CEST, ASL and Glucose-CEST 1 1 1 1 1 2 3 JA Wells *, JM O’Callaghan *, HE Holmes , B Siow , P Torrealdea, M Rega, S Richardson , M O’Neill , EC Collins , 1+ 1+ N Colgan , MF Lythgoe 1. UCL Centre for Advanced Biomedical Imaging, Division of Medicine and Institute of Child Health, University College London, UK 2. Eli Lilly & Co. Ltd, Erl Wood Manor, Windlesham, Surrey, GU20 6PH, U.K. 3. Eli Lilly and Company, Lilly Corporate Center, Indianapolis, IN 46285, USA *joint first author. + joint senior author. Introduction Transgenic mouse models of Alzheimer’s disease (AD) allow controlled assessment of the relationship between clinically relevant MRI parameters and underlying pathology. The TG4510 mouse model mimics several characteristics of AD through 1 over expression of mutant human tau and the development of neurofibrillary tangles (hyper-phosphorylated tau protein) . In this study we use arterial spin labelling (ASL), Chemical Exchange Saturation Transfer (CEST) and Glucose-CEST (a novel 2 technique to image tissue glucose uptake ) to investigate a TG4510 mouse model of AD and litter matched controls. Recent more effective treatments to control the abundance of tau protein demand new, safe, non-invasive methods to detect and quantify protein deposits in vivo. This is the first application of ASL and CEST to the TG4510 model and the first application of Glucose-CEST to investigate neurodegenerative disease. Methods 9 transgenic TG4510 and 17 wild-type (WT) litter matched control mice (8.5 months) were imaged using a 9.4T VNMRS horizontal bore scanner (Agilent Inc.). A 72mm inner diameter volume coil was used for RF transmission and signal was received using a 4 channel array head coil (Rapid Biomedical). Anaesthesia was induced and maintained at 1.5% isoflurane 3 in 100% O2 . ASL: A flow-sensitive alternating inversion recovery (FAIR) sequence with a 4-shot segmented spin-echo EPI readout was implemented with the following parameters: 5 slices, slice thickness =1mm, FOV=20x20mm, matrix size = 64x64, slice selective inversion pulse width =12mm, 5 inversion times (TI = 600, 1250, 1500, 2000, 2500 ms). CEST: Gradient 2 echo images (TR=6.1ms, TE=2ms, flip=5°, FOV=20×20mm , slice thickness 3mm, matrix size=64×64, line width <30 Hz) were acquired following a train of saturation pulses at 79 frequency offsets covering ±6 ppm to encompass APT saturation peaks around +3.5ppm. An estimate of ATP was calculated as the area under the MTRasym curves between 3.3 and 3.7ppm by subtracting the signal intensities at either side of the direct water saturation peak. Glucose CEST: Mice were fasted 24h before imaging. CEST measurements (using identical parameters described above), were applied at baseline and then every 8 minutes following an IP injection of 1g/Kg glucose for a total of 100 minutes following glucose delivery. (B) CEST Wildtype TG4510 (C) Glucose-CEST Wildtype TG4510 AreaMTRasym CBF (ml/100g/min) (A) ASL Wildtype TG4510 Figure 1 Mean CBF (A) and area under the MTRasym (B) within a ROI in the cortex of the TG4510 and WT mice. Each dot represents an individual animal. (C) The mean change in MTRasym from baseline flowing glucose injection across all TG4510 and wildtype animals. Results and Discussion Figure 1 (A,B) shows that the ASL and CEST measurements in the cortex completely discriminate between the TG4510 and litter matched control mice. This demonstrates the marked sensitivity of these techniques to tau pathology which may inform the development of a multi-parametric biomarker for clinical diagnosis. The increased CBF in the cortex may reflect 4 autoregulatory failure (isoflurane is a potent vasodilator) or compensatory processes associated with neuronal dysfunction . Figure 1 (C) demonstrates an increased mean glucose-CEST signal in the TG4510 mice. Although these results are not discriminatory between the groups, the increased mean glucose-CEST signal may indicate an accumulation of glucose in the extracellular space due to reduced cellular metabolism. Work is underway to investigate possible correlations of the MR measures to tau density assessed by histology. Our data suggest that these methods may serve as a platform for the assessment of novel therapies to halt or reverse disease progression. 1. Ramsden et al., J. Neurosci. 2005;25(46):10637–10647.2. Walker-Samuel et al., Nature Medicine 2013 in press 3. Kwong KK, et al Proc Nat Ac Sci 89:5675-5679 1992 4. Alsop DC, et al. 2008 Neuroimage Oct 1;42(4):1267-74. Epub 2008 Jun 17. Title: Connectivity effects of Ketamine its modulation by Risperidone and Lamotrigine 1 1 1 1 Authors: R. Joules , O.M Doyle , O. O’Daly , S. De Simoni , M.A. Mehta 1 1 Kings College London, Department of Neuroimaging, Centre for Neuroimaging Science (PO 89), Institute of Psychiatry, De Crespingy Park, London, SE5 8AF, UK. Background: The NMDA antagonist ketamine induces glutamatergic dysfunction in humans and has been linked to the transient induction of psychotomimetic symptoms. A more detailed understanding of the effects of ketamine on the brain will contribute to an appreciation of the role of glutamate in psychotomimetic symptoms and may inform the development and assessment of novel treatments. Indeed, the effects of ketamine on the blood oxygen level dependant (BOLD) signal in the brain have been widely studied and robust regional modulations with antipsychotics identified; however, its effects on whole-brain connectivity are only beginning to be studied. Aims: Here we investigate the effects of ketamine on regional coupling during resting-state and examine the modulatory effect of risperidone and lamotrigine pre-treatment on ketamine-induced connectivity patterns. We hypothesise ketamine will have a robust effect on connectivity within the brain and that pre-treatment with lamotrigine and risperidone will attenuate this connectivity state through direct and in-direct modulation of glutamate levels. Furthermore, we expect pre-treatment with lamotrigine and risperidone to differ in resultant network patterns, due to their differing mechanisms of action on glutamatergic function. Methods: Data were collected from sixteen healthy male volunteers over four sessions in a placebo controlled cross-over design experiment. These sessions comprised a control session and three sessions involving a ketamine infusion. In order to investigate the effects of glutamatergic and antipsychotic drugs on the ketamine response, two ketamine infusion sessions included an oral pre-treatment of either lamotrigine or risperidone. Given the distributed effects of glutamate on local and long-range transmission in the brain, we applied graphtheory centrality measures to assess the degree to which ketamine alters regional connectivity. These centrality measures were used as a feature set for multivariate group comparisons using Gaussian process classification (GPC) and leave one out cross validation (LOOCV) in order to investigate differing patterns of centrality across the whole brain. Results: We observed that ketamine robustly and predictably altered network centrality measures throughout the brain with respect to the placebo condition. Specifically, ketamine attenuated centrality in regions including the lateral, frontal and cingulate cortices whilst increasing the centrality indices for the parietal and occipital lobes, cerebellum and basal ganglia. Considering the ketamine infusion and saline infusion session both with placebo pre-treatment resulted in a significant classification of 81.25% (p=0.001). Risperidone treatment prior to infusion significantly affected centrality compared to placebo pre-treatment (Acc=75%, p=0.001) and pre-treatment with risperidone robustly modulated the centrality metrics observed for the subsequent ketamine infusion in comparison to the placebo pre-treated ketamine infusion (Acc=71.88%, p =0.003). Furthermore, following risperidone pre-treatment, ketamine-related centrality patterns did not significantly differ from those seen following placebo (saline) infusion (Acc=62.5%, p=0.076). The classifier was unable to accurately distinguish between lamotrigine and placebo pre-treated ketamine states (Acc=40.6%, p=0.922), but was able to significantly separate the saline condition from the ketamine state pre-treated with lamotrigine (Acc =75%, p=0.001). Conclusion: Here we demonstrate the effectiveness of using centrality measures for whole brain, multivariate group comparisons within a drug effect study and with possible applications to the study of psychiatric disorders. Our results indicate sub-anaesthetic doses of ketamine reduce connectivity between frontal regions and the rest of the brain whilst enhancing connectivity in subcortical and posterior cortex. Importantly, whilst lamotrigine which is known to modulate ketamine-induced glutamate levels, here, it did not significantly modulate ketamine-related regional connectivity; risperidone, was shown to attenuate the altered patterns of connectivity seen following ketamine administration. This suggests that the modulation of connectivity produced by ketamine may be primarily due to NMDA receptor blockade rather than increased glutamate levels. A novel method of minimizing EEG artefacts during simultaneous fMRI: a simulation study. M. E. H. Chowdhury, K. J. Mullinger, A. Antunes, P. M. Glover, R. Bowtell SPMMRC, School of Physics & Astronomy, University of Nottingham, Nottingham, NG7 2RD Introduction: The utility of EEG-fMRI is limited by the large artefact voltages that are produced in EEG recordings made during concurrent fMRI. Novel approaches for reducing the magnitude and variability of the artefacts are therefore required. One such approach involves using an EEG cap incorporating a reference layer (RL), which has similar conductivity to tissue and is electrically isolated from the scalp [1, 2]. The RL carries a set of electrodes and leads that precisely overlay those on the scalp, so that similar voltages are induced at the RL and scalp electrodes by time-varying magnetic fields (Fig. 1). RL Artefact Subtraction (RLAS), which involves taking the difference of the voltages at the two electrodes, should therefore attenuate artefacts whilst leaving neuronal voltages unaffected. Aim: Here we use electromagnetic modelling to simulate the voltages induced in a hemispherical RL and a spherical volume conductor (VC) by a time varying magnetic field gradient. By evaluating the differences in the voltages REF Electrode produced in the RL and VC as the RL geometry is varied, we test the efficacy of the RLAS Reference layer (e.g. agar-layer) approach and identify an optimal RL design. Insulating layer Methods: We modelled the head as a homogeneous spherical VC of 9cm radius, which was (e.g. PVC) GND Electrode separated from the hemispherical RL by a thin insulating layer (Fig. 1) and evaluated the peak voltages generated by a transverse (right-left) gradient of 10mTm-1 amplitude varying Reference Layer at 1kHz (dG/dt_peak = 62Tm-1s-1). A finite volume method (204×204×204 voxels, at 1mm Electrode resolution), based on a quasi-static approximation of Maxwell’s equations was used in conjunction with an analytic form of the driving vector potential. The resulting linear system Scalp Electrode Figure 1 Schematic representation of the RLAS system of equations was solved using the BiCGSTAB algorithm [3, 4]. experimental setup. Proposed electrode set-up on scalp and We assessed the difference in voltages induced on the surface of the two conductors at 33 reference layer with insulating layer separating the electrode locations, defined by the extended 10/20 system electrodes (insert). mV (VC at isocentre and electrode Cz positioned on z-axis) 8 B C A while the RL geometry, defined by the RL thickness 4 (RLT), insulating layer thickness (ILT) and RL angular o o 0 extent (RLAE) [90 =hemisphere; 180 =sphere], was varied. Parameters were varied sequentially with the -4 optimum value for earlier parameters set in each case with -8 RLT=5mm, ILT=1mm and RLAE=90o, initially: a. 1˚ or 2˚ mismatch in the electrode locations on the two conductors 170 ms - 810 ms b. RLAE=80o to 155o -276.94 µV 0 0.00 µV 139.11μV -277 139 µV c. RLT=2 to14mm Figure 2 Potential distribution due to a time-varying a transverse magnetic field for a hemid. ILT=1 to10mm spherical conducting RL (A) and spherical VC (B). C: Artefact map for the difference in For each case the voltages at the VC and RL electrodes induced voltages between the VC and RL for RLT=5mm, ILT=1mm and RLAE=90̊ . 1200 300 were found and the difference calculated. The RLT=5mm RLAE=87˚ B C 800 A 1000 250 ILT=1mm ILT=1mm RMS amplitudes of the voltage difference and 800 200 600 voltage at VC electrodes were also calculated. 600 150 400 400 100 Results: The strong similarity of the induced 4mm 1mm RLAE=87˚ 87˚ 200 200 50 potentials in the VC and RL, when a RLT=4mm 0 0 0 2 4 6 8 10 80 90 100 110 120 130 140 150 2 4 6 8 10 12 14 transverse gradient is applied, is evident (Fig. Insulation layer thickness (mm) Reference layer Angular extent (deg) Reference layer thickness (mm) 2A&B). The RMS of the potential induced on Figure 3 RMS over sample locations of the difference in induced voltages between the VC and RL for a range of the VC for the realistic initial parameters reference and insulation layer geometry parameters. Varying: A) RLAE; B) RLT and C) ILT. employed was 3233μV. Fig. 2C shows maps of the difference in induced voltages between the RL and VC (Fig. 2(B–A)), showing the large attenuation of the induced artefact by subtraction of the RL voltages (RMS=125μV). Given the close agreement between the RL and VC voltages it is unsurprising that a small mismatch between sample locations produces large difference in measured artefact voltages: for 1˚/2˚ of mismatch RMS=232/427V, demonstrating sampling locations must overlay precisely. Fig. 3 shows that the minimum discrepancy between induced voltages is achieved with RLAE=87˚ and that this is further reduced by reducing the RLT to 4mm. However, the ILT must be as small as possible (1mm) to give the minimum difference in induced voltages (RMS=34V). Conclusions: This work suggests the optimal set-up for an RLAS system would be RLT = 4mm, ILT = 1mm, RLAE = 87o. In this case the RMS difference in induced voltages between conductors was found to be 34V for an applied transverse gradient, a 91dB attenuation of the VC voltages. This simulation work suggests that RLAS is a viable method for removing artefacts in EEG data during simultaneous fMRI, and the RLAS approach should therefore be experimentally tested. 170 ms - 810 ms 0.00 µV 139.11 µV RMS Potential (V) RMS Potential (V) RMS Potential (V) -276.94 µV References 1. Dunseath, W.J.R., Apparatus and method for reducing interference 2008: US. 2. McGlone, F. and R. Dunseath, Localizing EEG abnormalities using neuroimaging. Epilepsy and Behavior, 2009. 16(1): p. 6-7. 3. Antunes A, et al., Magnetic field effects on the vestibular system: calculation of the pressure on the cupula due to ionic current-induced Lorentz force. Phys. Med. Biol., 2012. 57: p. 4477-4487. 4. Van Der Vorst, H.A., BI-CGSTAB: a fast and smoothly converging variant of BI-CG for the solution of nonsymmetric linear systems. SIAM J. Sci. Stat. Comput., 1992. 13(2): p. 631-644. BC-ISMRM XXII Postgraduate Symposium 2013 36 Abstracts: Oral Presentations Cancer, Cardiac & Nerves 37 Lactate and glutamine as potential MRS-detectable metabolic biomarkers of treatment response to the novel HSP90 inhibitor NVP-AUY922 in human breast cancer cells E Wholey, E de Billy, M O Leach, P Workman, M Beloueche-Babari Institute of Cancer Research, 15 Cotswold Road, Sutton, SM2 5NG Background: Progress in the development of targeted therapeutics requires clinically applicable pharmacodynamic (PD) biomarkers in order to assess target modulation and treatment response. Previous work has shown that inhibition of heat shock protein 90 (HSP90) – a molecular chaperone which stabilizes oncogenic proteins - led to promising anti-tumour activity in human cancer cells and tumour models that was associated with MRS-detectable changes in choline metabolism [1-3]. The aim of the current study is to assess the effects of HSP90 inhibition on additional key cancer metabolic pathways using the novel and clinically relevant HSP90 inhibitor NVP-AUY922. Methods: MDA-MB-231 and MCF7 human breast cancer cells were treated with NVP-AUY922 for 24 hours. MCF7 cells were also treated with the alternative HSP90 inhibitor 17AAG for 24 hours. HSP90 inhibition was monitored by assessing the expression levels of client proteins - CRAF and CDK4 - and the heat shock protein HSP72, which is induced upon HSP90 inhibition. Metabolite content was analyzed by 1H-MRS of cell extracts and cell culture media from control and treated cells and corrected for cell number. A B Results: Treatment with NVP-AUY922 resulted in a decrease in intra-cellular lactate: 31 ± 11% in MDA-MB-231 (n=4, p<0.001) and 34 ± 14% in MCF7 (n=5, p<0.001) compared to controls. [no MCF7 data] Extra-cellular lactate levels were also decreased following treatment in both cell lines (up to 46 ± 10%, n=3, p=0.001). Furthermore, significant changes in glutamine metabolism were detected, characterized by Figure 1 A Intra-cellular metabolic changes calculated from proton a 23 ± 5% increase in glutamine in MDA-MB- MR spectra following 24 hour NVP-AUY922 treatment in MDA-MB231 cells (p=<0.0001) and a 22 ± 7% decrease 231 and MCF7 human breast cancer cells. B Changes in lactate in glutamate in MCF7 cells (p=<0.0001) production and glucose consumption measured by 1H-MRS of cell compared to controls. Since glutamate is culture media samples. Data are expressed as a percentage of produced from glutamine, these observations control (mean ± SD) **p=0.001; ***p=<0.0001. may reflect altered activity of the glutamine transporter or the enzyme glutaminase. Similar results were observed in MCF7 cells following exposure to the alternative HSP90 inhibitor 17AAG. Lactate is produced primarily via glucose metabolism to pyruvate, which is converted to lactate via lactate dehydrogenase-A (LDH-A). To investigate the mechanism behind the fall in lactate, we assessed glucose uptake from the cell culture media by 1H-MRS, and the expression levels of LDH-A and other key glycolytic enzymes – specifically GLUT1 and hexokinase 2 (HK2) - by western blotting. Our data show a 48 ± 10% decrease in glucose consumption in MDA-MB-231 cells (n=3, p=0.001) and preliminary results indicate a reduction in GLUT1 and HK2 protein expression in both MDA-MB-231 and MCF7 suggesting reduction in glycolytic flux to pyruvate. LDH-A expression was unaltered. Discussion and conclusions: HSP90 inhibition has an effect on glycolysis and glutamine metabolism, which are both primary pathways of energy production in cancer cells. The observed decrease in both intra- and extra-cellular lactate was, in MDA-MB-231 cells, associated with a decrease in glucose consumption, suggesting reduction in overall glycolytic flux. Further work seeks to investigate the mechanism of the change in glutamine. These findings highlight the potential of lactate and glutamine as metabolic biomarkers of treatment response to HSP90 inhibitors. 1. Beloueche-Babari, M., et al. Oncotarget, 2010. 1(3): p. 185-97; 2. Chung, Y.L., et al. J Natl Cancer Inst, 2003.95(21): p. 1624-33. 3.Brandes, A.H., C.S. Ward, and S.M. Ronen. Breast Cancer Res, 2010. 12(5): p. R84. We acknowledge the support received from the CRUK and EPSRC Cancer Imaging Centre in association with the MRC and Department of Health (England) grant C1060/A10334, and NHS funding to the NIHR Biomedical Research Centre. This work was also supported by Cancer Research UK grant number C309/A11566 and the Rosetrees Trust. Diffusion Weighted MR imaging using an Active Breathing Coordinator to support Radiotherapy treatment planning E.Kaza, DJ.Collins, R.Symonds-Tayler, R.Panek, MO.Leach CR-UK and EPSRC Cancer Imaging Centre, Institute of Cancer Research, Sutton, Surrey, UK Background and aims: An Active Breathing Coordinator (ABC) (Elekta Oncology Systems, Crawley, UK) is a respiratory control apparatus employed during lung, liver or breast Radiotherapy (R/T) to reduce motion and treatment margins. This device monitors patient breathing by means of a propeller flow transducer and obstructs airflow at a preset air volume for a defined duration by inflation of a balloon valve. Our purpose is to acquire diffusion weighted (DW) Magnetic Resonance (MR) images in ABC-controlled breath holds (BH) as applied during R/T irradiation, which would be useful for treatment planning and response assessment, and to assess image quality compared to standard motion control techniques. Methods: A healthy volunteer was scanned supine with arms above his head in a 1.5T Siemens Avanto using an axially 2 oriented echo-planar imaging (EPI) sequence with fat saturation (SPAIR) and four b-values: 0, 100, 500, 750 s/mm in 3 2 scan trace mode. The sequence parameters were TR 1110ms, TE 67ms, matrix 128x104, FoV 360x360 mm , partition thickness 6mm, α 90°, GRAPPA factor 2. Imaging was performed initially in five 16s long self-sustained BH, guided by the operator. The same EPI sequence was repeated twice with ABC-induced BH, achieving the same volume of lung inflation during both measurements. When the volunteer was requested to take a deep breath through the ABC spirometer his inhaled air volume exceeded the preset threshold of 1.2l and the balloon valve was inflated. Upon detection of valve closure, a custom circuit triggered the sequence for a preset duration of up to 16s. The MR buzzer was modified to interrupt BH in case of an emergency. Then EPI DWI was also applied under free breathing using a navigator on the diaphragm with a +/-2.5mm acceptance window. Results: The ABC application caused neither volunteer discomfort nor affected the MR images; the ABC function was not affected by the static or gradient magnetic field. The same liver slice is depicted in figure 1 with the same 2 2 windowing for the self-controlled BH, one ABC-controlled BH Figure 1. DWI with a) b=0 s/mm , b) b=500 s/mm and c) ADC maps under I) self-controlled BH on inhalation, II) ABC-controlled BH on and the navigated free breathing sequence for b = 0 and 500 2 inhalation and III) navigated free breathing. s/mm . The apparent diffusion coefficient (ADC) map of these slices is also shown. A higher signal intensity is observed for BH imaging using the ABC. As the same lung inflation volume was achieved under ABC control for all BH during the two subsequent EPI acquisitions, the images were aggregated thus increasing the image signal to noise ratio (SNR). Improved contrast under ABC allows more accurate structure delineation. Figure 2. ADC maps of the kidneys for I) self-controlled BH on Figure 2 shows the ADC map of the kidneys equally inhalation, II) aggregated measurements of ABC-controlled BH on inhalation and III) navigated free breathing. windowed under each breathing condition, illustrating that the two ABC measurements aggregation yields a smoother ADC map than self-sustained BH and allows a better differentiation of the renal cortex from the medulla than navigated free breathing. Conclusions: An ABC apparatus provided improved DWI compared to the standard self-sustained BH and navigated free breathing methods in this study, under clinically applicable BH conditions. As images were acquired with the same lung volume and organ position during different acquisitions, they could be aggregated to improve structure delineation. The device can be used with clinically relevant imaging protocols to achieve the same organ positions required during R/T to inform treatment planning and assess irradiation effects. Any desired phase of the respiratory cycle can be blocked for an arbitrary duration. Our modified ABC allowed for automated starting and ending of MR scanning with the onset and termination of a BH. Volunteer safety was guaranteed by the modified MR buzzer which, if pressed, would immediately interrupt the ABC-sustained BH. Multi-modal MRI and FDG-PET for assessment of treatment in the infarcted mouse heart G Buonincontri1, C Methner2, T Krieg2, T A Carpenter1, S J Sawiak1,3 1 Wolfson Brain Imaging Centre, Department of Clinical Neurosciences, University of Cambridge, Cambridge, UK 2Department of Medicine, University of Cambridge, Cambridge, UK 3Behavioural and Clinical Neuroscience Institute, University of Cambridge, Cambridge, UK Purpose Chronic heart failure, as a result of acute myocardial infarction (MI), is a leading cause of mortality in the Western world. Novel treatments aim to prevent cell death in the ischaemic myocardium and the consequent development of heart failure. MRI is an excellent tool to investigate tissue viability with late gadolinium enhancement (LGE MRI), to monitor remodelling by measuring global function with cine sequences, as well as local muscle contractility with displacement encoding with stimulated echoes (DENSE MRI [1]). Imaging with MR allows accurate assessment of anatomy, motion and viability, however the signal is non-specific. Positron emission tomography (PET) is a complementary technique, which is highly specific for molecular imaging but lacks anatomical detail. Combining these techniques offers a sensitive, specific and quantitative tool for the assessment of new therapies. We performed cardiac PET/MRI sequentially in a preclinical study investigating a novel treatment. Methods We imaged mice (n=6 with a novel treatment; n=6 with a placebo) 24 hours after MI. We performed MRI with a Bruker BioSpec 47/40, using a 12 cm diameter birdcage transmitter and a 2 cm diameter surface loop receiver. Functional assessment of the left and right ventricle with cine MRI (FISP, TR/TE 7ms/2.4ms, 13-20 frames, 3.5 cm FOV, 256x256 matrix, 1 mm slice thickness, bandwidth 64kHz, flip angle 20°, NEX 2, prospectively gated), regional radial and circumferential strain with DENSE MRI [1] (3 short-axis slices encoding from end diastole to end-systole, 1 mm thick, TR/TE ~200 ms/9.5ms, 3.5 cm FOV, 128 matrix, bandwidth 64kHz, flip angle 90°, 4 NEX with CANSEL [2]), as well as tissue viability with LGE MRI [3] were performed. The imaging bed was then transferred from the magnet to the PET camera, leaving the receiver coil in place. After injection of 20 MBq of FDG, list mode data were acquired for 45 minutes to assess tissue metabolism. Results Coregistration between PET and MRI data could be performed easily using an axial translation. There was good correlation between MRI functional parameters, MRI-derived strains and MRI-derived infarct size and PET measurement of infarct size, which were each individually able to detect differences between different treatments (p<0.05, t test). PET infarct size correlated well with MRI, as areas of enhancement in the LGE MRI, corresponding also to areas of reduced muscle activity revealed by MRI-derived strain measurements, matched regions of reduced uptake in the PET images. Small and non-transmural infarct zones were not visible in the PET images, probably due to partial volume effects. Conclusions This study demonstrates a method for performing sequential PET-MRI which allows easy coregistration using standard equipment. Both MRI and FDG-PET were able to distinguish between different treatments, although non-transmural infarcted zones, clearly identified in MRI, were not detected by PET. The same technique can be used to investigate the uptake of novel PET tracers in the heart, where accurate coregistration is crucial to interpretation. Figure 1 Multi-modal assessment of a mouse heart 24 hours after induced MI. a) LGE-MRI, the infarct appears as an area of hyperenhancement. b) FDG PET, the infarct appears as an area of reduced uptake of FDG. c) Displacement map during systole obtained from DENSE, the arrows represent the movement of each pixel. Infarcted areas have impaired movement. d) Circumferential strain obtained from the displacement. The non-viable area has impaired contractility. [1] W Gilson et al 2005, Am J Phys Heart Circ Phys 288(3):H1491:7 [2] F H Epstein, W D Gilson 2004, Magn Reson Med. Oct;52(4):774-81. [3] G Buonincontri et al 2013, J Magn Reson Imaging 10.1002/jmri.24001 Comparison of Arterial Spin Labelling and R2* as Predictive Response Biomarkers for Vascular Targeting Agents in Liver Metastases S.P. Johnson 1,2*, R. Ramasawmy 2*, A. Campbell-Washburn2, M. Robson1, V. Rajkumar1, S. Walker-Samuel2, M.F. Lythgoe2, R.B. Pedley1 1 UCL Cancer Institute 2 UCL Centre for Advanced Biomedical Imaging *Joint first authors Introduction: Metastatic liver disease is the main cause of mortality in colorectal carcinoma (CRC) patients, with a 5 year survival rate of 40% following surgical resection of metastases1. Surgery with curative intent is only possible in 10-20%1 of patients, demonstrating the need for alternative therapeutic approaches. The vascular disrupting agent OXi4503 is a compound that targets tumour vasculature and causes central tumour necrosis2 leaving a small viable rim of tumour cells3. The acute (within 4 hours) accumulation of paramagnetic deoxyhaemoglobin resulting from vascular disruption has allowed R2* changes to be used as a biomarker of therapeutic effect 4. However, Arterial Spin Labelling (ASL) could offer an alternative quantifiable technique for assessing response, by measuring acute changes in tumour perfusion using wholly endogenous contrast mechanisms 5. The current study therefore aims to compare changes in R2* and ASL following OXi4503 treatment in a preclinical liver metastasis model. Method: Animal model: The CRC cell line SW1222 was injected intrasplenically at a concentration of 1x10 6 cells in 100 µl in serum free media into n=6 MF1 nu/nu mice. Cells were allowed to wash through to the liver for 1 minute followed by splenectomy. Solid tumour deposits developed within the liver at ≈4 weeks following surgery. MRI: A 9.4T Agilent VNMRS 20cm horizontal bore system with a 39mm birdcage coil was used, with a warm air blower to maintain animal temperature. Respiratory gating (SA instruments, New York, USA) was used on all scans. Fast spin echo images were used to define a suitable imaging slice within the liver followed by a segmented FAIR Look-Locker ASL sequence with a single slice spoiled gradient readout [5]. R2* values were assessed by a multigradient echo (MGE) image sequence covering the entire liver. FAIR Look-Locker ASL sequence parameters: 30 x 30mm FOV, 128x128 matrix, TE: 1.18 ms, TI: 110 ms, TRRF: 2.3 ms, TRI: 13 s, 50 inversion recovery readouts. Localised inversion thickness: 6 mm, imaging readout slice thickness: 1 mm, 4 lines per segmented acquisition. MGE sequence parameters: 8 echoes, TE1=2ms, echo spacing=2ms, TR=280ms; 128x128 matrix, 40x40mm FOV, 1mm slice thickness. Dosing: Cannulation of the tail vein was performed prior to baseline scans. Dosing of 40mg/kg via this remote i.v. line was performed in the scanner bore after baseline scans, and data acquired at 90min post dose. Data analysis: n=18 metastases were evaluable across the n=6 mice for ASL and n=12 were available for R2* analysis. Perfusion maps were generated using the Belle model7, (T1blood =1.9 s 8, blood-tissue partition coefficient λ=0.95 ml/g 9) in MATLAB and R2* maps were created using IDL. A B C Fig 1: Plots showing the liver metastases response 90 minutes post OXi4503 against baseline measurements for perfusion (A) and R2* (B). A significant trend can be seen in the perfusion changes compared to the initial perfusion, however no trend can be seen in the post dose vs baseline R2*. No significant correlation can be seen between the change in R2* and perfusion post dose (C). Results: A significant decrease was measured in ASL measurements of tumour perfusion at 90 mins following OXi4503 administration (P < 0.01, MannWhitney U test), with a mean change of -0.49 ml/g/min (-43%). A significant correlation was observed between baseline perfusion and the change in perfusion following therapy (Fig. 1A), suggesting that tumours better perfused at baseline responded better to the therapy. A significant increase in R2* was also measured (P < 0.01, Mann-Whitney U test), with a mean change of 0.010 ms -1 (13%), but with no significant correlation with initial R2* (Fig.1B). There was no significant correlation between ASL and the R2* responses (Fig.1C). Discussion: We were able to detect acute changes in tumour pathophysiology caused by OXi4503 with both ASL and R2*, with a significant decrease in mean perfusion and increase in R2*. This is consistent with the mechanism of action of VDAs: cessation of blood flow leads to a reduction in tumour perfusion and an increase in paramagnetic deoxygenated haemoglobin. Changes in R2* and perfusion were not correlated, indicating a complex relationship between changes in flow and accumulation of deoxyhaemoglobin, which may be specific to individual tumours. The data presented here shows that ASL can be a predictor of vascular targeting agent efficacy in liver metastases, suggesting that tumour deposits better perfused at baseline display a greater acute response. R2* response was not suggestive of any prognostic ability, but did respond positively. Given the mechanism of action of vascular disrupting agents, ASL provides response biomarkers that afford a less ambiguous interpretation than intrinsic susceptibility (R2*) measures. However, an approach combining the two may provide deeper insights in to the mechanics of tumour response in vivo, by relating flow changes to changes in blood oxygen saturation. The detection of a variable response, even in tumour deposits within the same liver highlights the need for robust assessment of response within individual patients. ASL sequences are non-invasive and do not require the administration of a contrast agent and so could be performed serially, soon after therapy to inform on drug efficacy. Given that brain and kidney FAIR ASL is commonplace in clinical scanners we anticipate a translation of hepatic ASL should be straightforward. Further work will characterise the response at later time points post OXI4503 and assess changes in perfusion and R2* in other tumour lines in preclinical metastases models. Acknowledgements: This work was carried out as part of King’s College London and UCL Comprehensive Cancer Imaging Centre CR-UK & EPSRC, in association with the MRC and DoH (England). We would also like to thank OXiGENE for supplying OXi4503. References: (1) Penna C and Nordlinger B, British Medical Bulletin (2002) 64:127-140, (2) Chan LS et al, Anticancer Drugs (2008) 19(1):17-22, (3) Pedley RB et al, Cancer Research (2001) 61(12):4716-22, (4) Zweifel M et al, Proc. Intl. Soc. Mag. Res. Med. 19 (2011) 339, (5) Ramasawmy R et al, Proc Brit Chap ISMRM (2011) poster 62, (6) Chan LS et al Anticancer Res (2007) 27:2317-2324. (7) Belle, et al. J Magn Reson Imaging 1998;8;12491245. (8) Campbell A, et al. Magn Reson Med. 2012; doi. 10.1002/mrm.24243. (9) Rice G et al, J Pharmacol Methods. 1989 Jul;21(4):287-97. Measurement of magnetization transfer effects in the Brachial Plexus: comparison with T2 and Diffusion effects Zaid Bin Mahbub1, Andrew Peters1, K Siddique Rabbani2, Olivier Mougin1, and Penny Gowland1 1 SPMMRC, School of Physics & Astronomy, University of Nottingham, Nottingham, United Kingdom, 2Department of Biomedical Physics and Technology, University of Dhaka, Bangladesh MTR or D (10-3 s/mm2 ) Background: Brachial nerve disorders can result from Cervical Spondyltic Radiculopathy (damage to nerves from vertebra (osteophyte)) or myelopathy (damage within spinal cord). Both lead to similar functional symptoms and signs on EMG. MRI provides the opportunity to obtain more precise information about the location of the injury and quantitative biomarkers of the injury. However tracking the nerves and nerve roots can be difficult because of the complexity of the anatomy, making it hard to perform quantitative imaging in the area. Diffusion weighted whole body imaging with background suppression (DWIBS) has been previously proposed as a method of highlighting the nerve roots separately from background tissue [1]. Aims: to combine DWIBS with magnetization transfer preparation, varying b value and diffusion time diffusion encoding, and varying echo time to measure magnetization transfer ratio (MTR), diffusion time dependent diffusion coefficient (D) and T2 in the brachial plexus. Methods: Scanning was performed according to local ethics committee approval on 4 subjects aged 33-49 y.o. (2 female) with no history of injury to the neck or arm. All Figure 1 Coronal MIP of brachial plexus, MTR (off res = 0.4 scanning was performed on a Philips 3T Achieva scanner using the torso 16 kHz), T2 and Diffusion map (Δ = 18.3ms) of the spinal cord channel array coil. The basic sequence used for all measurements was an and nerve roots. inversion recovery (TI=400 ms for fat suppression), pulsed gradient 2 spin echo, single shot EPI sequence, (3mm isotropic resolution, MT(Cord) 1.8 192×60×300mm FOV, fat-water shift=4.64 pixels), 18 transverse1.6 oblique slices centred on C5-C6) generally with TE= 100 ms and D_Δ1(Cord) diffusion encoding to suppress the signal from all tissues except the 1.4 D_Δ2(Cord) nerves to assist in image analysis. For MTR this was preceded by a 1.2 o train of 8 300 flip angle MT pulses played out at 20 ms intervals; the MT(Nerve) 1 MT scan was repeated 9 times with off-resonance of ±1000, ±600 D_Δ1(Nerve) 0.8 (optimal to detect NOE effect in nerves), ±400 and 0 Hz and finally no off resonance pulses, TR=6s, b=500 s/mm2, Δ=28.3, δ=10ms. For T2 0.6 D_Δ2(Nerve) 6 different echo times were acquired: TE=55,60,65,70,75,80 ms, 0.4 2 TR=6s, b=500 s/mm , Δ=81.3ms, δ=10ms. For diffusion 0.2 measurements data was acquired for b=300,600 s/mm2, with diffusion 0 times Δ1=18.3 and Δ2=81.3ms, δ=10ms, TE= 100ms, to give 0 200 400 600 800 sensitivity to restricted diffusion (b=0 not used due to IVIM T2(ms) contamination). Coronal MIPS were created to confirm the location of Figure 2 Relationship between D, MTR and T2 values from both sides the brachial plexus. Sagittal images were reconstructed through the nerves and cords for all subjects (Δ1=18.3ms, Δ2=81.3ms). nerve roots. ROIs were selected over the nerve roots at C5/C6/C7/C8 Table1: MT, T2 and D values of cord and nerves and over the spinal cord, automatically based on their high image MT T2(ms) DΔ1(mm²/s) DΔ2(mm²/s) contrast by fitting with 2D Gaussian surface typically one voxel for each nerve location and five voxels for cord location. T2 and D were Nerve 0.30±0.08 129±39 1.43±0.32 1.40±0.27 calculated using a linear fit to log(Signal) versus TE and b. MTR was Cord 0.33±0.03 458±181 1.07±0.22 1.15±0.36 calculated from the difference between the saturated and non saturated signal, normalized to the saturated signal (only data for offset of ±600 Hz reported). Results: Figure 1 shows a coronal MIP through the brachial plexus and MTR, T2 and ADC maps, Figure 2 shows the relationship between MTR, D and T2 measured in healthy volunteers for both the nerves and cord. There was a negative correlation between D and T2 measurements for cord and nerves (R² = 0.87, 0.12 for Δ1 and R² = 0.67, 0.23 for Δ2 respectively). The intersubject averages are summarized in Table 1. Across all subjects diffusion measured with Δ2 tended to be lower than that measured with Δ1. Conclusion: A protocol has been developed to allow the measurement of D, T2 and MT in the brachial nerves, with limited contamination from CSF or blood flow by the inclusion of diffusion weighting in the sequences. In this healthy group no particular trends are expected in the data. However, figure 2 suggests a negative correlation between D and T2, which seems to be similar across both nerve and cord. This is contrary to what might be expected due to CSF contamination (although the wide range of T2 values measured in the cord do suggest some csf contamination there). Alternatively this may reflect weighting of signal to different tissue compartments as T2 changes. The close correlation observed between independent measurements of D for different diffusion times indicate real biological variation between subjects. The tendency for D81<D18 suggests an effect of restricted diffusion is seen in the nerve. This set of sequences combining DWIBS with different quantitative imaging techniques provides method for comprehensive quantitative evaluation of the brachial plexus and it will now be applied to clinical groups. References: [1]Takahara et al, Radiat Med 22:275–82. Acknowledgment: This work was funded by the Islamic Development Bank (IDB). Abstracts: Posters and Poster Pitches 43 Design of trapezoidal oscillating gradients for diffusion MRI A.Ianus1, B.Siow1,2, I. Drobnjak1, H. Zhang1, D. C. Alexander1 1 Centre for Medical Image Computing, 2Centre for Advanced Biomedical Imaging, University College London Introduction: The current work presents the benefits of trapezoidal oscillating gradient spin echo (OGSE) sequences for temporal diffusion spectroscopy, where only sinusoidal waveforms have been considered previously [1]. This approach studies the diffusion properties in the temporal domain, and OGSE sequences are used to probe diffusion over different time scales. Thus, in the case of restricted diffusion, it is possible to infer microstructural properties of the underlying tissue from the dependence of the estimated apparent diffusion coefficient (ADC) on oscillation frequency. We show how to design the trapezoidal waveforms when the slew rate of the gradient is taken into account, in order to avoid sampling long diffusion times, which might corrupt the analysis. Theory: In a temporal diffusion spectroscopy experiment, oscillating Fig.1: Schematic representation of a) pulsed gradient spin echo (PGSE), b) square OGSE and c) trapezoidal OGSE gradients are used to measure the diffusion spectrum at the same frequency as the oscillation. As illustrated in Fig. 1, OGSE sequences are parametrized by the gradient amplitude, G, the duration of the pulse, δ, the interval between the onset of the first and the second gradient, Δ, the frequency of the wave, ω, the phase, φ, and, in the case of the trapezoidal wave, the rise time, tr. The signal attenuation due to free diffusion of water molecules can be written as: S ( 2τ )=exp ( −b⋅D ) where 2τ is the echo time, b is the diffusion weighting factor and D is the diffusion coefficient. As described 2 T in [2], b can be calculated using a modified form of the Bloch equations: b=γ F ( t ) F ( t ) dt, , where γ is the gyromagnetic ratio of t the hydrogen nuclei and F ( t ) =∫0 G ( t' ) dt' with G(t) the effective field gradient. When working in the temporal domain, the diffusion signal γ2 ∞ can be written as S =exp − ∫0 F ' (ω)⋅D ' (ω) F ' ( ω) where 2 D'(ω) is the temporal spectrum of the molecular motion and ( 2τ a) b) ) i ωt F ' (ω)=∫0 e F (t )dt is the power modulation spectrum [3], which indicates the diffusion times contributing to the signal. Fig.2: a) Example of effective To sample a single component of D'(ω) it is essential that c) gradient waveforms with ω=718 F'(ω) has a single peak at the desired frequency. For this reason, Hz and an appropriate choice of cosine-like waveforms, i.e. φ=π, are preferred over sine-like Δ-δ; b) Corresponding power waveforms, i.e. φ=π,, as they do not have a peak at ω=0. However, modulation spectra; c) Power modulation spectra when Δ-δ is the instant gradient increase of a cosine or square waveform is increased by half a period. unachievable in practice. To overcome this problem, [1] introduced b) the so called apodised cosine, in which the first oscillation is a sine a) function with double frequency. The current work shows how to construct apodised trapezoidal waveforms. Simulations and results: For trapezoidal waveforms, the trivial case when the duration of the first oscillation is a quarter of a period, exhibits a secondary peak in the power modulation spectrum at ω=0. Thus, we construct apodised trapezoidal waveforms with the desired properties by extending the duration of the first and last oscillations by half the rise time, tr/2. Moreover, we study the Fig.3: a) Diffusion signal for different waveforms as a function of oscillation frequency for four different sizes of the restricted compartment; influence of the time interval between the 1st and 2nd gradient (Δ-δ) corresponding extracted ADC values. The diffusion signal and ADC for apodised on the power modulation spectrum. Δ-δ should be carefully chosen s.t. b) trapezoid and square wave are very similar and are plotted on top of each other. the effective gradients (including the effect of the refocusing pulse) are as close as possible to a continuous waveform. Otherwise, there is a split in the main peak of F(ω), Fig 2c. As a proof of concept, we used synthesised diffusion MRI data for three different cosine-like oscillating waveforms (square, apodised cosine and apodised trapezoid) to investigate the power modulation spectra and the dependence of the signal and estimated ADC on frequency. The signal is computed using Callaghan's matrix method [4,5] for a two compartment tissue model which mimics white matter [6]: S ( 2 τ)= f ⋅S r +(1− f )⋅S h , where f is the volume fraction of axons, Sr is the signal from the cylindrically restricted compartment and Sh is the signal from the hindered compartment which exhibits Gaussian diffusion with parallel diffusivity Dr and perpendicular Dh. The parameter settings used in the simulation are: −9 2 −9 2 G=0.1 T/m , δ=35 ms , Δ=40 ms , and ω∈π {1,2,... , 16}. We f =0.7, Dr =1.7×10 m /s , D h =1.2×10 m /s , ADC =log (S ( 2 τ))/ b on oscillation frequency for different radii of the cylinder investigated the dependence of the estimated compartment R∈{1,2,5,10 }μ m and the results are illustrated in Fig. 3a and 3b. The peak heights in Fig. 2b. are similar for square and apodised trapezoid and lower for the apodised cosine, indicating less focus on the desired diffusion time. The ADC values are similar for all waveforms, nevertheless square and trapezoidal waves provide higher diffusion weighting. Discussion: We show that square and trapezoidal waveforms are beneficial for temporal diffusion spectroscopy, as they yield higher diffusion weighting per oscillation compared to sinusoidal oscillations, leading to improved signal-to-noise ratio. With the correct apodisation, trapezoidal waveforms have the desired properties of cosine-like function for studying the dependence of ADC on frequency. The application of square and trapezoidal waveforms for techniques that explicitly model diffusion restriction, such as ActiveAx [7], is extensively discussed in [8]. References: [1] Does MRM03, [2] Price 97, [3] Callaghan JMRA95, [4] Callaghan JMR95, [5] Drobnjak JMR11, [6] Panagiotaki NI12, [7] Alexander MRM08, [8] Ianus JMR12 Novel, unsupervised method for iron deposit segmentation in the basal ganglia A. Glatz1,2; M.C. Valdés Hernandéz1-3; A.J. Kiker1; M.E. Bastin1-3; I.J. Deary3,4; J.M. Wardlaw1-3 BRIC, 2SINAPSE, 3Centre for Cognitive Ageing and Cognitive Epidemiology, 4Department of Psychology; University of Edinburgh Background and Aims: Basal ganglia iron deposits (BGIDs), which mainly arise from ferrungination of small vessels and perivascular spaces, typically appear as focal hypointensities (Fig. 1A) in and around the globus pallidus on T2*-weighted MRI of elderly, community-dwelling subjects [1]. Their manual segmentation is generally labour intensive and results have a high inter-rater variability. Additionally, blooming artefacts and partial volume effects limit the accurate volumetry of BGIDs on T2*-weighted MRI, where calcifications can also be confounding features. Here we present an unsupervised, automated BGID segmentation method, which addresses the aforementioned difficulties. The method is intended for the application in large-scale and inter-centre studies, since further research into the biological mechanisms of BGIDs is needed and could possibly lead to a new MRI biomarker [1]. Methods: Iron deposits, such as BGIDs, generally appear iso- and hypointense on T1- and T2*-weighted MRI volumes [2]. As BGIDs are typically very small features BGID signal intensities typically resemble outliers in joint signal intensity scatter plots (Fig. 1B). For their segmentation co-registered T1- and T2*-weighted volumes were masked with basal ganglia nuclei masks and model based clustering was applied to reduce segmentation and imaging artefacts. Clusters with T1- and T2*-weighted intensities from normal-appearing basal ganglia tissue were then used to estimate the robust intensity means and covariance matrices [3]. The Mahalanobis distance of each T1- and T2*-weighted intensity pair of a basal ganglia nucleus from the corresponding normalappearing tissue intensity distribution was measured. T1- and T2*-weighted intensities were classified as outliers if the Mahalanobis distances exceeded an adaptive threshold [4]. The final BGID masks were obtained after filtering preliminary outlier masks with morphological operations to reduce booming artefacts and remove potential calcifications. The BGID segmentation method was implemented in Matlab and validated with co-registered clinical T1- and T2*-weighted volumes from 65 elderly, otherwise healthy subjects in their 70s [5]. The basal ganglia masks were automatically generated with FSL FIRST [6]. The similarity between BGID masks from the presented segmentation method and manually created masks from an experienced rater was assessed with the Jaccard similarity index and Dice coefficient [7]. Furthermore, the similarity of the total BGID volumes of the masks from the presented method and the rater (Vtotref) was analyzed with Bland-Altman plots. The spatial coincidence of the BGID masks from all subjects generated with the method was assessed with spatial probability density maps [8]. 1 Results: The Jaccard similarity indices and Dice coefficients were 0.62±0.11 and 0.77±0.12 with the masks from the presented BGID segmentation method and the rater. These figures indicate that the masks from the BGID segmentation method were in moderate to substantial agreement with the masks from the rater. The method segmented larger BGIDs better than smaller BGIDs, since the Bland-Altman plot showed a higher variability for smaller (Vtotref<40mm3) than for larger BGIDs. The Bland-Altman plot also revealed a slight proportional error, since the masks from the method for subjects with Vtotref<90mm3 were slightly larger compared to the ones from the experienced rater and vice-versa. The spatial probability density maps (Fig. 2), which was constructed with all BGID masks from the method, confirmed that the highest spatial coincidence of all BGID masks also coincided with the region, where the penetrating, potentially iron encrusted arteries enter the globus pallidus [9]. Conclusions: The presented BGID segmentation method can generate BGID masks that are similar to masks from an experienced rater. The high variability in the Bland-Altman plot for subjects with smaller BGIDs is, at least partly, due to the facts that (i) small BGIDs (<20voxels in our case) reduces the statistical power of the method significantly and (ii) every false negative or positive voxel has a disproportional large impact on the similarity metrics. Higher (in-plane) resolution MR volumes could potentially improve the agreement between the rater and the method. References: [1] Penke, L. et al. Neurobiol Aging 2012 33(3):510-17. [2] Valdés Hernández, M.C. et al. Eur Radiol 2012 22(11):2371-81. [3] Verboven, S. et al. Chemometr Intell Lab 2005 75(2):127-36. [4] Filzmoser, P. et al. Comput Geosci 2005 31(5):579-87. [5] Wardlaw, J.M. et al. Int J Stroke 2011 6(6):547-59. [6] Patenaude, B. et al. NeuroImage 2011 56(3):907-22. [7] Shattuck, D.W. et al. NeuroImage 2009 45(2):431-39. [8] Glatz A. et al. NeuroImage subm. [9] Morris, C.M. Acta Anat (Basel) 1992 144(3):235-57. The utility of synthetic data for quantitative assessment of a DCE-MRI registration algorithm Anita Banerji1,2, Alexandra Morgan1,2,3, Yvonne Watson1,2, Giovanni A Buonaccorsi1,2,3, and Geoff J M Parker1,2,3 2 Centre for Imaging Sciences, University of Manchester, Manchester Academic Health Science Centre, Manchester, UK. 2 Biomedical Imaging Institute, Manchester, UK. 3 Bioxydyn Ltd, Manchester, UK Introduction Fitting a tracer kinetic model such as the extended Kety model1 to dynamic contrast-enhanced (DCE) MRI data allows estimation of parameters such as Ktrans that reflect microvascular characteristics. Prior to model fitting, a registration algorithm can be applied to reduce motion corruption of the signal intensity time course data for each voxel. However, it is hard to ascertain from in-vivo data whether the registration has increased the accuracy and precision of the estimated model parameters or whether model fitting itself is robust to motion. In contrast, synthetic data can be generated from known ground truth model parameters to allow quantitative assessment of the sensitivity of DCE-MRI parameterisation to motion and the potential benefit of registration. In this work we present a synthetic DCE-MRI time series of a liver tumour with breathing motion emulation and demonstrate the utility of the synthetic data by using it to assess a model driven registration algorithm2. Synthetic data We have developed a flexible software phantom generator3 and implemented the modules required to generate liver tumour DCE-MRI time series. Contrast agent time courses (75 time-points, temporal resolution of 4.97 s) were generated using the extended Kety model for the tumour core and rim and a dual-input model4 for the liver. A population arterial input function5 (AIF) was used as input to both models with a portal input function estimated from the AIF6 for the liver model. Dynamic T1 values were generated using the known relationship with contrast agent concentration and relaxivity. Signal intensity time courses were then simulated using the spoiled gradient echo pulse sequence equation (TR = 4 ms, TE = 0.82 ms, flip angle = 30°). The ground truth values for the tracer kinetic model parameters were based on 6 patient liver metastases data sets where the modelling had been applied, giving tumour core and rim Ktrans values of 0.18 min-1 and 0.36 min-1 respectively. The anatomy was generated from masks defined on an end-exhale high resolution contrast-enhanced CT data set from a single individual. Motion was emulated by applying displacement maps from a finite element modelling based registration procedure7 to the end-exhale masks. A breathing trace, acquired using respiratory bellows during a dynamic MR acquisition, was used to determine the percentage of displacement to apply at each time point. The generated images were downsampled to a resolution of 2.9 x 2.9 x 4 mm3, which produced partial volume effects. Zero mean Gaussian noise with an SNR of 7 was added to the signal. Two synthetic data sets were produced: motion-free and motion-corrupted (see Fig. 1). 30s 35 s 40 s 45 s 50 s 55 s 6 min 8 s Figure 1. Time-points 7 to 12 and 75 from the synthetic data set with motion emulation. The tumour core is indicated by the arrow. Data analysis Model driven registration2 was applied to the motion-corrupted data set using translation only deformations. The extended Kety model was then fitted to motion-free, motion-corrupted and registered data sets on a per voxel basis. The AIF and portal input function as described above were used as input to the model fitting process. Results Median Ktrans for the tumour region is 0.31 min-1 for motion-free, 0.29 min-1 for motioncorrupted and 0.31 min-1 for registered data sets (see Fig. 2). No significant differences were found between the median values of the data sets using the Mann-Whitney U test. Ktrans histograms for the tumour region are shown in Fig. 3. A bi-modal distribution is seen in the motion-free and registered data set but not the motion-corrupted data set. Conclusion Knowledge of the ground truth data from which the synthetic images were generated trans allows us to demonstrate that, for this data set, the median tumour Ktrans value is robust to motion and Figure 2. K box plots for the tumour region. The registration recovers of the bi-modal distribution of the tumour. As the flexible phantom generator central line is the median software application used to generate the synthetic data may be of value to other researchers in this value. field we have made it available for download at http://www.qbi-lab.org/software. References [1] Tofts, Mag Res Im 7:91 1997. [2] Buonaccorsi, Mag Res Med 58:1010 2007. [3] Banerji, Proc ISMRM 16:493 2008. [4] Materne, Clin Sci (Lond) 99:517 2000. [5] Parker, Mag Res Med 56:993 2006. [6] Banerji A, Mag Res Im, 35:196 2012. [7] Brock, Int J Oncol Biol Phys 64:1245 2006. Acknowledgements We thank Kristy Brock, Princess Margaret Hospital, University Health Network, University of Toronto, ON, Canada for the CT data set with organ masks and displacement maps and Tim Cootes, Imaging Sciences, University of Manchester, UK, for guidance on software libraries for applying the displacement maps. Figure 3. Ktrans histograms for the tumour region for the motion-free, motioncorrupted and registered data sets. MRI of epilepsy therapy: Dexamethasone may exacerbate cerebral oedema in a rat model of status epilepticus. Duffy B A1, Scott R C2 3, Lythgoe M F1 1 Centre for Advanced Biomedical Imaging (CABI), Department of Medicine and Institute of Child Health, University College London (UCL), UK 2 Department of Neurology, Neuroscience Centre at Dartmouth, Dartmouth Medical School, Lebanon 03756, NH, USA 3 Institute of Child Health, UCL, UK Background and Aims: Some studies have shown various anti-inflammatory drugs to be neuroprotective or anti-epileptogenic following status epilepticus (prolonged seizures). For example, Fabene et al. showed that blockade of leukocyte-endothelial interactions following status epilepticus (SE) via administration of α4 integrin specific antibodies, reduced the occurrence of spontaneous seizures in the chronic epileptic phase[1]. The mechanism of action for these drugs is thought to occur via protection of the blood brain barrier (BBB). In order for these findings to be translated into a clinical setting, there needs to be a biomarker for therapy monitoring. T 2 weighted MRI can be used as a biomarker of vasogenic oedema. In this study we test the hypothesis that dexamethasone reduces T2 at 2 or 4 days following pilocarpine induce SE in rats. Methods: Animal Model: Lithium chloride (3 meq/kg, i.p.) was administered to Sprague-Dawley rats. Rats were then pretreated with methyl scopolamine nitrate in order to reduce the peripheral effects of pilocarpine. 30 min later, Pilocarpine hydrochloride (30 mg/kg, i.p.) was given in order to induce status epilepticus (n=16). Control animals did not receive pilocarpine. Animals were behaviourally scored on the Racine seizure severity scale. SE was terminated after 90 min using diazepam. Following SE, animals were randomly assigned to 2 groups: A control group which received saline injections (SE) and a therapy group which received a single dose of dexamethasone sodium phosphate (SE-DEX) (2 mg/kg, i.p.). MRI Imaging: MRI was performed using a 9.4 Tesla DirectDrive VNMRS horizontal bore scanner with shielded gradient system (Agilent Technologies, Palo Alto, CA) and a 4-channel rat head phased-array coil (Rapid Biomedical GmbH, Würzburg, Germany).T2 measurements were performed across 15 contiguous slices using a multi-slice multi-echo spin-echo sequence using a TR = 2.5 s and the following echo times: TE = 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120 ms, FOV = 25 x 25 mm, slice thickness = 1 mm, matrix = 128 x 128. Analysis: Regions of interest were identified by rigid co-registration of the multi-echo images to a rat brain MRI template. For each region, the mean value at each echo time was used to fit a single exponential decay using nonlinear least squares regression. Statistical analysis was performed using Mann-Whitney U-tests. Results: There was no difference between the two groups in seizure severity according to the behavioural scoring or latency to seizure onset. 2 of the 6 animals (33%) in the SE-DEX group died within 24 h of SE. The cause of death was unknown but was unlikely to be due to the diuretic effect of dexamethasone as saline was administered to replace lost fluids. Hippocampus: At 2 d the T2 was significantly elevated in the SE-DEX group compared to controls (p=0.03) but this was not significant in the SE group (p = 0.07). Piriform cortex: At 2d the T2 in both SE and SE-DEX rats were elevated (p = 0.01 and p = 0.03 respectively). Furthermore, the T2 appeared to be longer in the piriform cortex in rats treated with DEX compared to untreated rats (p = 0.06). Thalamus: T2 changes were limited to the lateral posterior thalamic nucleus and the dorsal lateral geniculate nucleus. T2 measurements began to normalise across all brain regions at 4 d following SE indicating that these alterations are transient. Due to an as yet unexplained reason, 2 rats in the SE group showed marked unilateral T2 alterations in the neocortex that were not present in any of the other rats. Conclusions: There is reasonable evidence to suggest that even in low doses, such as those used in the current study; dexamethasone exacerbates the regional cerebral oedema that ensues transiently following prolonged seizures in the rat. The mechanisms by which this might occur are unknown. References: [1] P. F. Fabene et al., Nature Medicine 14, 1377 (2008). Agreement in Reproducibility of Whole-brain Structural Connectivity Networks with Alternative Pipelines CS Parker1,2, CA Clark1, S Ourselin2 and JD Clayden1 Imaging and Biophysics Unit, Institute of Child Health, UCL, 2Centre for Medical Image Computing, UCL 1 Background Mapping whole-brain structural connectivity using diffusion-weighted imaging has recently become a popular research topic in neuroscience. Here, the brain is considered as a network of white matter tracts linking grey matter regions. Connectivity between all regions is quantified using tractography techniques and network properties, describing the local and global topology of anatomical connections, can be derived using graph theoretical analysis [1]. Previous studies have shown that network properties such as global clustering coefficient and pathlength, are altered in neurological disease, meaning such network metrics may be useful biomarkers [2]. However, a variety of different methods may be used to reconstruct the structural network and there is no agreement on which method gives metrics of the highest anatomical relevance or is most sensitive to disease [3,4]. To address this issue, the reproducibility of network metrics obtained from two different reconstruction pipelines was investigated; one pipeline used commonly available segmentation, registration and tractography tools whereas the other used tools more recently developed at CMIC. Objectives This work aims to compare the intra and inter-subject reproducibility of whole-brain network metrics derived using two different reconstruction pipelines. Methods 2 T1-weighted images (3D FLASH) and 3 repeats of a 63-direction echo-planar diffusion weighted sequences were acquired at 1.5T from a group of 28 young healthy subjects. Two pipelines (N and F are delineated using '/') were applied to reconstruct whole-brain connectivity. The structural images were skull-stripped, registered to improve signal:noise, and parcellated into 44/68 cortical regions, as defined in the Hammers/Desikan Atlas, using NiftySeg/Freesurfer software. Structural images were non-linearly registered to the subjects b=0 image using NiftyReg/FSL registration tools. The voxel orientations of white matter fibers in diffusion-weighted images of (2.5mm) 3 resolution were obtained using Constrained Spherical Deconvolution (CSD)/FSL-BEDPOSTX. 100 probabilistic streamlines were seeded from each grey matter boundary voxel, and connectivity was quantified between all node pairs as the sum of connecting streamlines divided by the mean node boundary volumes. In a preliminary analysis, we investigated the reproducibility of the raw connection matrices (CM) using the intra- and inter-subject coefficient of variation (CV) and intra-class correlation coefficient (ICC). Results Accuracy of cortical parcellations, registrations, fiber modelling and streamline tracking were reasonable as assessed by visual inspection. Using the native atlas node resolutions, N-reconstructed CMs had higher subject grand mean connection strengths (0.08±0.18 vs 0.03±0.09), intra- and inter-subject reproducibility (CV, intra-: 0.44±0.23 vs 0.73±0.39, inter- 1.10±0.43 vs 2.02±0.93) and ICCs (0.72±0.16 vs 0.62±0.19) than F-reconstructed networks. (Fig. 1, left). To remove the possible effect of node scale on connection strength and reproducibility [6], nodes in N and F atlases were merged, based on nomenclature, yielding a common consensus node scale of 34. N-34 and F-34 networks had higher mean reproducibility than their higher resolution counterparts (mean ICC of 0.73±0.16 and 0.68±0.17, Fig. 1, left) and the trend of higher reproducibility for N reconstructed networks remained (Fig. 1, right). Figure 1. (Left). Intra-class correlation coefficients (ICC) of all connections in N-X and F-X pipelines, where X describes the number of network nodes. (Right) Difference in ICCs of all connections in N-34 and F-34 networks. Red and blue indicate higher reproducibility in the N-34 and F-34 pipelines, respectively. Conclusions Identifying and reducing sources of variation in network connections will allow greater discriminatory power for clinical studies of network architecture. T hese results support recent work showing higher reproducibility of the N pipeline using an alternative tractography scheme [7] and agree with previous studies suggesting that measurement variation in network connection strengths may be dependent on the reconstruction method used [2,3]. N networks had higher densities and connection strengths, which may reflect the greater ability of streamlines to track through areas of complex fiber architectures using CSD, or a superior delineation and alignment of grey matter nodes to diffusion space. Connections in N networks had lower intra and inter-subject CVs and high ICCs, even when the effect of node scale was removed, suggesting that this pipeline is generally less sensitive to noise and other image artefacts in repeat scanning and may therefore be better able to describe true subject anatomy. The current framework provides a basis for building reproducible networks and for examining consensus in reproducibility and connectivity between alternative pipelines whilst fixing the number of nodes [6]. Future work will examine the reproducibility of graph theoretical measures derived from thresholded connection strength matrices, aswell as the consensus and disagreement between N-34 and F-34 pipelines. References [1] Bullmore, E. & Sporns, O. 2009. Nat. Rev. Neuroscience. 10,186-198 [2] Shu, N., et. al. 2011. Cerebral Cortex. 11, 2565-77. [3] Bassett, D. et al. 2010. Neuroimage. 54, 1262-1279 [4] Cheng, H. et. al. 2012. Neuroimage. 61, 1153-1164 [6] Camoun, L. et al. 2012 Neuroimage. 203, 386-297 [7] Parker, C. et al. 2012. Proc. of BC-ISMRM, Cambridge Fitting the two-compartment filtration model in renal DCE-MRI by linear inversion D. Flouri†∗ , D. Lesnic† , S. Sourbron∗ Division of Medical Physics (∗ ) and Department of Applied Mathematics († ), University of Leeds INTRODUCTION: Improvements in DCE-MRI data quality have led to the development of increasingly complex tissue models, such as the 4-parameter two-compartment renal filtration model (2CFM) [1]. A practical limitation of these models is the need for non-linear least squares fitting. This is prohibitively slow for pixelby-pixel analysis, and requires a choice of initial values which may bias the results. The aim of this study is to develop a linear approach to fitting the 2CFM and use simulations to compare it to the non-linear method. THEORY: The 2CFM produces 4 parameters: the plasma volume (VP ), glomerular filtration rate (FT ), plasma flow (FP ) and tubular volume (VT ). The analytical solution provides a relation between the concentrationtime curves in renal tissue (C(t)) and a feeding artery (CA (t)): VT −t/TP VP −t/TP −t/TT e + e ⊗e ⊗ CA (t) (1) C(t) = TP TP TT Here TP = VP /FP and TT = VT /FT are the mean transit times of plasma and tubuli, and ⊗ is convolution. The standard non-linear model fit determines values for VP , VT , TP , TT by minimising the mean-squared difference between left and right hand side. To derive a linear inversion method we compute first and second time derivatives of C(t) from the model equations, and combine those into a single second-order linear partial differential equation: −C(t) 1 1 dC(t) VP dCA (t) VT + VP d2 C (2) + − + + CA (t) = 2 (t) TP TT TP TT dt TP dt TP TT dt The need for numerical differentiation is removed by integrating the equation twice over time. With N time points this forms a matrix equation AX = B with an N × 4 matrix A and an N -element array B of known concentrations, and a 4-element array X of unknowns. After solving for X using standard linear least-squares methods, TP , TT , VP , VT can be derived analytically. SIMULATIONS: Normal renal tissue (VP = 0.24, TP = 6.5s, VT = 0.63, TT = 125s) was simulated in IDL on a laptop PC. C(t) was calculated using Eq. 1 at a pseudo-continuous temporal resolution (0.1s), using a literature-based CA (t) [2]. To simulate measurement, C(t) and CA (t) were then downsampled at TR= 1s intervals for 5min, where the initial sampling point was randomly and uniformly distributed over the interval [0,TR]. No noise was added at this point. Non-linear fitting was performed using the IDL function CURVEFIT(), with initial values about half the exact values (VP = 0.1, TP = 3s, VT = 0.3, TT = 60s). Linear fitting was performed using the singular value decomposition of A. All simulations were repeated 10000 times to determine accuracy and precision. RESULTS: Values for the linear method are narrowly distributed around the exact values (Fig 1); histograms for the non-linear method are bi-modal, with one peak near the initial values, and a second near the exact value. This indicates that in a significant proportion of cases, the non-linear routine converged to a local minimum near the initial values. Calculation times for a typical MR slice of 128x128 pixels were 0.4hrs for the linear method, and 12hrs for the non-linear. Figure 1: Histograms for TP , TT , VP , VT (from left to right) with linear (blue) and non-linear (red) methods. Median values and 90% confidence interval are given in the table below. CONCLUSION: The linear fitting algorithm as proposed in this study reduces the calculation times, removes the bias caused by the choice of initial values, and increases the precision in each parameter. Future studies will verify whether these conclusions remain valid in the presence of noise, at lower temporal resolution, and in application to real data. REFERENCES: [1] Sourbron SP et al 2008 Invest Radiol 43: 40-8. [2] Parker GJM et al 2005 Magn Reson Med 56: 993-1000. ACKNOWLEDGEMENTS: This study was funded by an EPSRC-GSK CASE studentship. Optimised encoding scheme for vessel-selective arterial spin labelling 1 1 1 Eleanor S K Berry , Peter Jezzard and Thomas W Okell 1 FMRIB Centre, Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, United Kingdom Background and Aims: Vessel-encoded arterial spin labelling (VE-ASL) allows the blood from individual arteries to be traced in the brain. Typical VE-ASL schemes produce an approximately sinusoidal variation in inversion efficiency across space, allowing certain combinations of arteries to be separately tagged, or encoded. Combining information from a number of different encodings allows vascular territories [1] (Fig. 1a) and the brain vasculature [2] (Fig. 1b) to be mapped. The encoding schemes suggested thus far are only suitable for a small number of arteries, require manual prescription and may not optimise the SNR of the resulting images. Here we present an automated method that produces optimal encodings for an arbitrary set of vessels. a b Fig. 1. a: Vascular territories found using the standard method (axial slices). b: Dynamic angiogram of the Circle of Willis [2]. Right carotid = red, left carotid = green, right vertebral = blue, left vertebral = pink. Theory: The key to this method is to define the ideal encoding scheme, where combinations of arteries are perfectly tagged (-1) or controlled (+1), and then find the real encodings that best match this. We achieve this for each desired encoding by: 1. Constructing an “image” of the vessel locations. Each vessel is represented by +1 or -1, depending on the desired encoding (e.g. Fig. 2a) 2. Taking the Fourier transform of this “image”, masking and weighting it to up-weight lower spatial frequencies (Fig. 2b) 3. Finding the maximum intensity in this weighted Fourier space. This maximum point tells us the spatial frequency and phase to apply to best match our desired ideal encoding (Fig. 2c). Penalising high spatial frequencies from being chosen makes this method more robust to gross subject motion. Methods: The optimised method was compared to the simple scheme currently used for encoding the four main brain-feeding arteries (the right and left internal carotid and two vertebral arteries). The encodings are: non-selective tag and control, two leftright, two anterior-posterior and two diagonal. In this simple method left-right encodings are chosen to perfectly tag/control the c a b carotid arteries and anterior-posterior to Left carotid Right carotid tag/control at the average internal carotid/vertebral artery positions. Diagonal encoding attempts to perfectly tag the right carotid and left vertebral arteries, whilst Right vertebral Left vertebral controlling the other two vessels, or vice versa. The encodings produced by the two methods were simulated in MatLab for four Fig. 2. a: Idealised neck vessels showing the desired encoding pattern. b: vessels rotated by a range of angles about the Weighted and masked Fourier space, the position of the maximum intensity -1 centre. The resulting encodings were used to point is shown as a white pixel at k = (0.0156,-0.0156)mm . c: Neck vessels construct an “encoding matrix”. One measure of superimposed on the optimised spatial frequency; the maxima and minima performance obtained from this matrix is the of the spatial frequency line up well with the desired encoding pattern. overall encoding SNR efficiency (the higher the better, ideally = 1) [1]. Initial tests were done with the new method to see how it performs with more than four vessels, in this case on a set of vessels immediately above the Circle of Willis. The source of ideal encoding schemes for more than four vessels was the columns of a Hadamard matrix, which will lead to encodings with optimal SNR [1]. Results: When the four vessels are rotated beyond 30° the SNR efficiency of the standard encoding scheme deviates considerably from the ideal (average SNR efficiency = 0.57). However, the new method maintains near-ideal performance (average SNR efficiency = 0.99). Fig. 3 gives an example of how the spatial frequencies found with the new method fit a set of vessels above the Circle of Willis. The SNR efficiency of the complete encoding matrix is 0.63. Fig. 3. An encoding of Conclusions: The proposed method produces near-optimum vessel-encodings in the simple four vessel vessels above the Circle case, irrespective of rotation of the neck. Additionally the new method is capable of finding spatial of Willis. frequencies that match well the desired encodings of vessels above the Circle of Willis. However, experimental data is required to validate the proposed method and determine whether deviations from the assumed sinusoidal encoding function have a significant impact. In future work we plan to test the method on more realistic vessel patterns with four or more arteries. This could lead to more efficient encoding of a large number of vessels above the Circle of Willis; currently there is no optimised method for encoding these. Another comparison to be made is with random encoding [3], which requires less planning but theoretically has lower SNR efficiency and requires a greater number of encoding cycles than the proposed method. This becomes particularly important when the scan following each encoding takes time to acquire, as for vessel-encoded dynamic angiography [2]. References: [1] Wong 2007, Magn Reson Med 58:1086; [2] Okell 2010, Magn Reson Med 64:698; [3] Wong 2012, Magn Reson Mater Phy 25:95 Assessment of cardiac timing intervals using high temporal resolution real-time spiral phase contrast with UNFOLD-SENSE Grzegorz Tomasz Kowalik MSc1, Jennifer Anne Steeden PhD1, Oliver Tann MBBS1,2, Freddy Odille PhD3,4, David Atkinson PhD5, Andrew Taylor MD1,2, Vivek Muthurangu MD(res)1,2 1 UCL Institute of Cardiovascular Science, Centre for Cardiovascular Imaging, London, United Kingdom; 2Cardiorespiratory Unit, Great Ormond Street Hospital for Children, London, UK; 3IADI, INSERM U947, Nancy, France; 4Université de Lorraine, Nancy, France; 5Centre for Medical Imaging, UCL Division of Medicine, London, United Kingdom. Introduction: Cardiac time intervals are important cardiac health markers1, which can be as low as ≈30ms. Their accurate assessment require very high temporal resolution measurement of blood flow. We propose use of highly undersampled real-time spiral phase contrast MR (PC-MR) acquisition2 combined with parallel imaging (SENSE)3 and temporal encoding (UNFOLD)4. Real-time acquisition allows on assessment of each cardiac cycle individually. Non-Cartesian trajectories combined with parallel imaging reconstructions can provide resolution of 40-50ms2,5. However, for the sufficient resolution a further acceleration in form of temporal encoding is needed. In this study, a spiral UNFOLD-SENSE real-time PC-MR sequence was implemented with resolution of ≈13ms. Methods: All imaging was performed on a 1.5 Tesla MR scanner (Avanto, Siemens Medical Solutions, Erlangen, Germany) using 12 coils. Real-time data were acquired using a uniform density spiral PC-MR sequence2 with 12 interleaves required for fully sampled k-space (FOV: 500x500mm, matrix: 128x128, voxel size: 3.9x3.9x7mm, TR/TE: 6.58/1.97ms, flip angle: 15o). A total acceleration factor of 12 was used, resulting in one interleave per time frame and a temporal resolution of ≈13ms. A batched acquisition scheme was used to allow on calculation of coil sensitivity maps from data itself6 and use of alternating trajectories for temporal encoding. In which frames were acquired in blocks of 20; with each block containing a different set of alternating interleaves. The data from six blocks provides a fully sampled k-space for coil sensitivity calculation. The temporal filtering was performed in k-f space independently on each block, prior to the SENSE reconstruction. The alternating sampling pattern resulted in a twofold undersampling of each k-space position (Fig. 2). Subsequent Fourier transformation (FT) along time produced aliasing around the Nyquist frequency, which then was removed with a low-pass filter. The inverse FT of the filtered Fig. 2. Schematic visualisation of the temporal filtering signal is fully sampled and results in each frame containing two interleaves. used in the implemented UNFOLD-SENSE Additionally, to suppress possible ringing artefact, prior to FT each block was reconstruction. extended in both directions by replicating the first and last two frames twice. These were discarded after UNFOLD. The subsequent SENSE was performed as previously described7, in batches of six UNFOLD blocks. Two separate in-vitro experiments were performed to validate the UNFOLD-SENSE real-time PC-MR. Velocity quantification was compared to a retrospectively gated, reference standard Cartesian PC-MR sequence (FOV: 320x240mm, matrix: 256x192, voxel size: 1.3x3.9x5mm, TR/TE: 10.1/2.18ms, temporal resolution: ≈10ms, flip angle: 30o). Fifteen experiments were performed at different stroke volumes (30 to 65 ml), pulse rates (65 to 105 beats-per-minute) and peak mean velocities (30 and 65cm/s). Assessment of time interval measurements were done against simultaneous Fig. 1. Plot of LVOT and MVI velocity pressure measurements performed at the same position as the imaging planes using MR curves. Tangent lines drawn on the ascending compatible pressure transducers (Datex-Ohmeda, GE healthcare, Helsinki, Finland). Ten (+) and descending (-) slopes are used to experiments were carried out with temporal displacement of wave forms between 30 and calculate the cardiac time intervals. 200ms. Additionally, 10 healthy volunteers (all male of median age: 31.5 and range: 21-44 years) were scanned using the new sequence at rest and during exercise. Both the mitral valve inflow (MVI) and left ventricular outflow tract (LVOT) were imaged in a single imaging plain. At both stages, flow data were collected for ≈6.32s (480 continuous frames). The MVI and LVOT were manually segmented and mean velocity curves were extracted (Fig. 1). Subjective image scoring of rest and exercise data was performed by two independent, experienced observers who were presented with the magnitude images in a blinded, randomized manner. Results: There was no significant (P = 0.196) degradation of subjective image quality during exercise. Image segmentation and extraction of mean velocity curves were feasible for all data sets. Reconstruction time for 480 real-time frames, measured from the end of acquisition to data being available for viewing on a scanner console, was ≈23.45s. The in-vitro studies showed good agreement between the reference standard gated and real-time sequences (correlation coefficient: 0.99, P < 0.0001 and bias: -0.1cm/s, limits: -2.0 to 1.9 cm/s). Also, there was good agreement with the reference standard pressure measurement for timing assessment validation (correlation coefficient: 0.99, P < 0.0001 and bias: -0.7ms, limits: -6.9 to 5.4ms). Calculated parameters and results from in-vivo study are presented in Tab. 1. Discussion: We have shown that it is possible to measure the cardiac time intervals with very high temporal resolution real-time PC-MR. Our approach combined UNFOLD, SENSE and spiral k-space filling, which resulted in a temporal resolution of ≈13ms. This is a significantly higher resolution than previously described5,8. Using this sequence, it was possible to evaluate clinically important parameters at rest and during exercise for the first time using MR. We believe in the future this sort of approach may enable better assessment of patients with cardiac disease. HR [bpm] IRT [ms] ICT [ms] DT [ms] ET [ms] E/A Tei Index References: 1. Oh JK, et al. J Am 66.9±11.3 66.8±8.1 55.9±21.3 185.4±59.7 276.8±20.1 2.61±0.87 0.45±0.08 Rest Coll Cardiol 2003;42(8):1471Stress 87.6±11.3 57.9±9.5 42.6±19.5 104.0±32.5 264.8±21.4 1.95±0.58 0.38±0.09 1474; 2. Steeden JA, et al. J Magn Reson Imaging 2010;31(4):9972.8E-05 0.0213 0.0307 0.0001 0.0067 0.0017 0.0339 P 1003; 3. Pruessmann KP, et al. Tab. 1. In-vivo results (mean ± standard deviation and P values of paired two-tailed t-test) of the Magn Reson Med 1999;42(5):952cardiac timing parameters. 962; 4. Madore B, et al. Magn Reson Med 1999;42(5):813-828; 5. Joseph AA, et al. NMR Biomed 2012;25(7):917-924; 6. Kellman P, et al. Magn Reson Med 2001;45(5):846-852; 7. Kowalik GT, et al. J Magn Reson Imaging 2012; 8. Steeden JA, et al. Magn Reson Med 2010;64(6):1664-1670. The role of fronto-‐parietal networks in mental imagery Authors: Henrietta Howells, Flavio Dell’Acqua, Anoushka Leslie, Andrew Simmons, Christine Ecker, Mitul Mehta, Marco Catani, Declan G. Murphy, Michel Thiebaut de Schotten Introduction: When comparing two identical objects oriented differently, subjects rotate a mental image of one of the objects until it is congruent with the other. Evidence for this comes from the time taken to make the comparison, which increases linearly according to the extent of rotation needed to match the two objects (Shepherd, 1971). However this linear correlation is only evident in those using a visuospatial strategy to perform the task (Zacks, 2000). Some subjects may rely on a verbal strategy and this is reflected in a non-‐linear response time. A recent diffusion weighted imaging (DWI) tractography study reported a strong correlation between the hemispheric lateralisation of the white matter connections supporting the communication between the frontal and parietal lobe (the three branches of the superior longitudinal fasciculus – SLF I, II and III) and tasks involving visuospatial processing (Thiebaut de Schotten, 2011). However it is unknown whether this lateralisation influences the spatial manipulation of mental images. Here we explore the relationship between the lateralisation of the three branches of the SLF and performance in a mental rotation task. Methods: Twenty five healthy right-‐handed subjects (M:F 12:13, aged 22-‐35 years) were scanned on a 3T GE MRI scanner, (voxel size 2.4x2.4x2.4mm, 128&128 matrix, field of view 307x307mm, 60 slices, b-‐value 3000s/mm2, 60 diffusion-‐weighted directions and 7 non diffusion-‐weighted volumes). After correction for motion and eddy current distortions, whole brain spherical deconvolution tractography was applied for each diffusion dataset using Startrack (www.natbrainlab.com). Virtual in-‐vivo dissections of the three branches of the SLF were performed in each hemisphere and a lateralisation index calculated for the volume of each tract. Hindrance Modulated Oriented Anisotropy (HMOA), a new measure of microstructural properties of the tract was also extracted and a lateralisation index calculated (Dell’Acqua, in press). The subjects performed 128 trials of a computer-‐based mental rotation task using their right hand, with response time recorded using Superlab (Ecker, 2006). Mental rotation performance for each subject was determined using linear regression, looking at the subject response time dependent on the angle of rotation between the two objects. Bivariate analysis was performed to explore the correlation between lateralisation of the SLF and mental rotation performance. Results: Of the three branches, a significant positive correlation was found between right lateralisation of the SLF2 HMOA and a longer response time for the mental rotation task (2 outliers removed; RH: r=0.586, p<0.005). Conclusions: We found a correlation between the speed of mental rotation and the lateralisation of the second branch of the SLF. Our results suggest that having a more complex microstructure of the SLF2 in the right hemisphere would lead subjects to choose a visuospatial strategy to perform mental rotation. References: Dell’acqua, F. (in press), ‘Can spherical deconvolution provide more information than fiber orientations? Hindrance modulated orientational anisotropy, a true-‐tract specific index to characterize white matter diffusion’ Human Brain Mapping Ecker, C., et al. (2006), ‘Time-‐resolved fMRI of mental rotation revisited – dissociating visual perception from mental rotation in female subjects’ Neuroimage, vol. 32, no. 1, pp. 432-‐444 Thiebaut de Schotten, M. et al. (2011) ‘A lateralized brain network for visuospatial attention’ Nature Neuroscience, vol. 14, no. 10, pp. 1245–6 Shepard, R. (1971) ‘Mental Rotation of Three-‐Dimensional Objects’ Science, vol. 171, no. 3972, pp. 701-‐703 Zacks, J. (2000) ‘Mental spatial transformations of objects and perspective’ Spatial Cognition and Computation, vol. 2, 315–332 Tensor-based morphometry as a sensitive biomarker of Alzheimer’s disease neuropathology in a Tau transgenic mouse (rTg4510) 1* 1,2* 1 1 1 1 1 3 HE Holmes , NM Powell , JA Wells , JM O’Callaghan , N Colgan , B Siow , S Richardson , M O’Neill , 4 2 2 5 2 1 E Catherine Collins , J Cardoso , M Modat , E Fisher , S Ourselin, MF Lythgoe 1 UCL Centre for Advanced Biomedical Imaging, Division of Medicine and Institute of Child Health, UCL, UK 2 Centre for Medical Image Computing, UCL, UK 3 Eli Lilly & Co. Ltd, Erl Wood Manor, Windlesham, Surrey, GU20 6PH, U.K. 4 Eli Lilly and Company, 355 E Merrill Street, Dock 48,Indianapolis, IN 46225, USA 5 Department of Neurodegenerative Disease, Institute of Neurology, UCL, UK *Joint first author Introduction Alzheimer’s disease (AD) is a devastating neurodegenerative disorder, characterized by structural brain changes and cognitive impairment. The key neuropathological hallmarks of the disease are deposits of hyperphosphorylated tau and amyloid beta. The presence of these toxic aggregates is believed to precede a clinical diagnosis of AD by up to 15 years [1]. It is crucial that sensitive biomarkers of AD pathology are developed, to aid early diagnosis of AD and facilitate drug development. In this work, we have used an optimised sequence for high resolution in vivo μMRI to evaluate structural changes in the TG4510 mouse model of tauopathy and age-matched wildtype controls. This study will build on previous work [2] with higher field strength, optimised scan parameters and high resolution isotropic voxels, providing a novel platform for high sensitivity to subtle change in morphometry [3]. A new tensor-based morphometry (TBM) pipeline has been employed to locate regions of significant atrophy in the transgenic population. TBM’s value lies in its ability to detect change in any region of the brain without timeconsuming manual intervention such as the delineation of regions of interest. Methods Animals. Transgenic TG4510 mice and wild-type (WT) littermates were bred as published previously [4]. 9 TG4510 and 17 WT litter matched control mice (8.5 months) were imaged in vivo. Prior to imaging, mice were secured in a cradle under anaesthesia with 1-2% isofluorine in o 100% oxygen using a custom-built head holder to reduce motion. Body temperature was maintained at 36 – 37.5 C using a water-heating system and warm air fan. Core body temperature and respiratory rate were monitored using a temperature probe and pressure pad (SA Instruments, NY). Image acquisition. All scans were performed on an Agilent 9.4 T VNMRS 20 cm horizontal-bore system (Agilent Inc. Palo Alto, CA, USA). A 72 mm birdcage radiofrequency (RF) coil was used for RF transmission and a quadrature mouse brain surface coil (RAPID, Germany) was used for signal detection. A T2 weighted, 3D fast spin-echo sequence was implemented for structural imaging with the following parameters: FOV = 19.2 mm x 16.8 mm x 12.0 mm; resolution = 150 μm x 150 μm x 150 μm; TR = 2500 ms, TEeff = 43 ms, ETL = 4; NSA = 1. Total imaging time was approx. 1 h and 30 mins. Image processing. Images were brought into alignment via their principal axes, corrected for nonuniformity using an iterative expectation maximisation algorithm [5], and intensity-normalised. The images were masked using a STAPLE procedure which registered 10 sets of labels from the MRM NeAT mouse brain atlas to each image [6, 7]. Dilated masks and a target image from the population were used for both affine (5 iterations) and non-rigid (20 iterations) registration using the open-source NiftyReg package [8], to create an average atlas and deformation maps. Regions of significant volumetric change were found using the general linear model to apply a voxel-wise t-test on the log of the determinant of the Jacobian of these deformations, with animal weight used as a covariate. Results Figure 1: Representative axial, sagittal and coronal slices showing TBM results overlaid on an average structural image after 20 iterations of non-rigid registration. Figure 1 provides information on regions with a statistically significant (p < 0.05) local volume difference between groups. Regions of significant decrease (red) and increase (blue) in volume in the transgenic animals relatve to the controls are shown. We detected atrophy in both the hippocampus and the anterior cortex. These observations correlate well with the neuroanatomical regions known to be affected in both this mouse model and clinical cases of the disease [4]. In addition, TBM detected dilation in the lateral, third and fourth ventricles. Surprisingly, we also observed some local decrease in volume in the cerebellum, a region previously not known to be affected. This warrants further evaluation to confirm a volumetric change, via histology and both manual and automatic segmentation. Discussion We have used an automated pipeline for morphometric analysis which has detected significant volumetric changes between a transgenic mouse model of tauopathy and a control group. The detection of changes in regions expected for this model allows us to validate the use of TBM as an unbiased automated tool for the discovery of previously unknown changes, which we have noted in the cerebellum. We illustrate the potential of TBM as a sensitive biomarker for in vivo assessment of AD neuropathology in the mouse. Future work will validate the sensitivity of this method in a longitudinal study at earlier time points, where pathology is less severe. We intend to publicly release our TBM pipeline in the coming months. [1] Jack, C. R., M. A. Bernstein, et al. (2008) Journal of Magnetic Resonance Imaging 27(4): 685-691. [2] Yang, D., Z. Xie, et al. (2011) NeuroImage 54(4): 26522658. [3] Holmes, HE et al, BC ISMRM 2012. [4] SantaCruz, K., J. Lewis, et al. (2005) Science 309(5733): 476-481. [5] Cardoso, M.J., et al. (2011). NeuroImage, 56(3), 1386–97. [6] Magnetic Resonance Microimaging Neurological Atlas, "MRM NeAt", available from: http://brainatlas.mbi.ufl.edu. [7] Da M., et al. (Proceedings) MICCAI 2012 Grand Challenge and Workshop on Multi-Atlas Labeling. [8] Modat, M., et al. (2010). Comp. methods and programs in biomed., 98(3), 278–84. Low repeatability of BOLD during emotion processing is not due to physiological noise Ilona Lipp1,2, Kevin Murphy1, Xavier Caseras1,2, Richard Wise1 1 Cardiff University Brain Research Imaging Centre (CUBRIC), School of Psychology, Cardiff University 2 Institute of Psychological Medicine and Clinical Neurosciences, School of Medicine, Cardiff University Introduction: Previous studies have reported low repeatability of BOLD activation measures during emotion processing tasks (e.g. Plichta et al., 2012). It is not clear, however, whether low repeatability is a result of changes in the underlying neural signal over time, or due to insufficient reliability of the acquired BOLD signal caused by noise contamination. The aim of this study was to investigate the influence of “cleaning” the BOLD on measures of repeatability, by correcting for physiological noise and for differences in vascular reactivity. Methods: Fifteen healthy volunteers were scanned on two different occasions, performing an emotion perception task with faces (neutral, 50% fearful, 100% fearful) followed by a breath-hold paradigm. BOLD responses during the emotion paradigm were analyzed without and without physiological noise correction of the BOLD fMRI timeseries data. This correction consisted of: first applying correction of cardiac and respiratory artifacts (RETROICOR, Glover et al. 2000), followed by factoring out the influence of carbon dioxide (PetCO 2) level, oxygen (PetO2) level, heart rate (HR) and respiratory volume per time (RVT; Birn et al., 2006). The breath-hold task was used to obtain a vascular reactivity map for each participant, by using the recorded endtidal CO2 trace as a regressor in the GLM. Spatial repeatability and repeatability of signal amplitude within two regions of interest (amygdala, and fusiform gyrus) were estimated by calculating the intraclass correlation coefficient (ICC). Repeatability analysis was done on the dataset with and without physiological noise correction, and with and without accounting for vascular reactivity. Results: Significant repeatability of signal amplitude was found within the left amygdala during the perception of 100% fearful faces, none of the other ICCs reached significance. Spatial repeatability was higher within the fusiform gyrus than within the amygdala, and better on group level than on participant level. Neither physiological noise correction or considering vascular reactivity increased repeatability. Discussion: Repeatability of the BOLD signal during emotion processing could not be increased by correcting for physiological noise or accounting for day-to-day differences in vascular reactivity. These findings suggest that low repeatability is due to a lack of stability in the underlying neural signal. Further research is needed to identify fluctuating factors that influence neural activity and that could be used as regressors in order to obtain more repeatable activation measures. References: Birn, R. M., Diamond, J. B., Smith, M., Bandettini, P., 2006. Separating respiratory-variation-related fluctuations from neuronal activity-related fluctuations in fMRI. NeuroImage 31, 1536-1548. Glover,G.H., Li,T.Q., Ress, D., 2000. Image-based method for retrospective correction of physiological motion effects in fMRI: RETROICOR. Magnetic Resonance in Medicine 44 (1), 162-167. Plichta, M. M., Schwarz, A. J., Grimm, O., Morgen, K., Mier, D., Haddad, L., Gerdes, A. B. M., Sauer, C., Tost, H., Esslinger, C., Colman, P., Wilson, F., Kirsch, P., Meyer-Lindenberg, A., 2012. Test-retest reliability of evoked BOLD signals from a cognitive-emotive fMRI test battery. NeuroImage 60 (3), 1746-1758. In vivo Diffusion Tensor Imaging is sensitive to microstructural changes in both white and grey matter in the TG4510 mouse model of Alzheimer’s disease J O’Callaghan1, J Wells1, H E Holmes1, N Colgan1, B Siow1, S Richardson1, M J O’Neill2, E C Collins3, M F Lythgoe1 Centre for Advanced Biomedical Imaging, University College London1,Eli Lilly & Co. Ltd, Erl Wood Manor, Windlesham, Surrey, GU20 6PH, U.K 2. Eli Lilly and Company, 355 E Merrill Street, Dock 48,Indianapolis, IN 46225, USA 3 Aims We present Diffusion Tensor Imaging (DTI) as an in vivo imaging methodology sensitive to microstructural changes in a mouse model exhibiting tau pathology found in Alzheimer’s Disease (AD). Refinement of a biomarker sensitive to this pathology may lead to non invasive staging of AD that would have a massive impact on disease management and assessment of therapeutic pharmaceuticals. Introduction One of the pathological hallmarks of AD is the presence of intracellular neurofibrillary tangles (NFT) of aggregated hyper-phosphorylated Tau protein. These NFTs are found primarily to affect the structure of Grey Matter (GM) although early abnormalities in White Matter (WM) have also been observed in mouse models of AD1. DTI is an imaging technique sensitive to diffusion of water that can be used to probe tissue microstructure. Changes in mean diffusivity (MD) and fractional anisotropy (FA) in both white matter and grey matter have been observed clinically 2,3 in AD patients as well as in beta-amyloid pathology mouse models of AD4. In this study, DTI data was acquired in vivo for the TG4510 mouse, a model that exhibits selectivity tau pathology, along with age matched wild type controls. Identification of WM and GM regional differences between the groups may suggest sensitivity to microstructural changes caused by NFTs. Methods 9 Transgenic Tg4510 mice and 17 wild types were imaged in vivo at age 8.5 months. Mice were anaesthetized using 2% isoflurane and 1 L/m O2 and positioned in a custom-built head holder at which point the levels were reduced to 1.5% and 0.5 L/m respectively for imaging. A four shot Spin Echo EPI sequence was used to acquire sixteen slices on a 9.4T Agilent scanner using Rapid Biomedical RF coils (72mm volume coil Tx / four channel array head coil Rx). The olfactory bulbs were used as an anatomical landmark to maintain consistency in slice positioning between animals. The in plane resolution was 200 x 200µm with a slice thickness of 0.5mm. Diffusion gradients were applied in thirty directions with the following parameters G=0.25 T/m, Δ = 9.3ms, δ = 5.5ms, and b= 1050 s/mm2. Acquisition of 5 averages with a TR of 2000ms gave a total imaging time of 45 minutes. Removal of data corrupted by imaging artefacts reduced the group sizes to n=7 for post processing. In house software was used to construct tensors at each voxel through a least squares solution approach5. MD, FA, and principle eigenvectors were calculated from the tensors. A slice was selected in the unweighted images corresponding to a position 0.6mm posterior to bregma. Regions of interest were drawn manually on this slice in the corpus callosum and the in the forelimb section of the somatosensory area of the cortex (S1FL) for each animal. Mean values were calculated for the ROIs in each group (±1SEM) for comparison of the DTI parameters. Results A large decrease in FA and a small increase in MD were observed in the corpus callosum of the transgenic group as shown in Figures 1a and 1b. The direction of greatest diffusion was consistent between the groups. In the forelimb area of the cortex, there was a small increase in FA in the transgenic group in addition to a change in the direction of greatest diffusion from the rostral-caudal(B) to the medial-lateral direction(R), as shown in Figures 1c and 1d. Discussion At the time point imaged, the tau pathology in this model is fairly advanced with high numbers of tangles found in the Cortex and hippocampus accompanied with severe forebrain atrophy caused by neurodegeneration. The reduction in FA coupled with an increase in MD in the corpus callosum may be due to degradation of WM integrity. In the cortex, the increase in FA and change in direction of the principle direction of diffusion may be due to two possible causes, the restriction of diffusion caused by large amount of NFTs and structural changes in tissue caused by cell death. Conclusions The data presented shows that in vivo DTI is sensitive to both GM and WM changes caused by tau pathology at late stages of disease in the TG4510 mouse. Further work will test the sensitivity of this method by imaging mice at an earlier time point with less severe pathology and by comparison with a histological analysis of the imaged mice. References 1 Desai, M. K. et al. Glia 57, 54-65, doi:10.1002/glia.20734 (2009),2 Rose, S. E. et al. Journal of Neurology, Neurosurgery & Psychiatry 69, 528-530, doi:10.1136/jnnp.69.4.528 (2000). 3 Jacobs, H. I. L. et al. Alzheimer's & Dementia, doi:10.1016/j.jalz.2011.11.004, 4 Sun, S.-W. et al. Experimental Neurology 191, 77-85, doi:10.1016/j.expneurol.2004.09.006 (2005). 5 Batchelor, P. G., Atkinson, D., Hill, D. L. G., Calamante, F. & Connelly, A. Magnetic Resonance in Medicine 49, 1143-1151, doi:10.1002/mrm.10491 (2003). Acknowledgements This work is funded by Eli Lilly & Co and the UK Engineering and Physical Sciences Research Council Construction of a 4-Channel Transmit Neck Array for pCASL Tagging at 7 Tesla and Comparison with a Head Coil. Konstantinos Papoutsis1, James A Meakin1, Aaron T Hess2, Jamie Near3, Stephen J Payne4, David Edwards4 and Peter Jezzard1 1 FMRIB Centre, Nuffield Department of Neurosciences, University of Oxford, Oxford, UK. 2Department of Cardiovascular Medicine, John Radcliffe Hospital, University of Oxford, Oxford, UK. 3Centre d'Imagerie Cérébrale, Douglas Institute, McGill University, Montreal, Canada. 4Department of Engineering Science, University of Oxford, Oxford, UK. Target Audience: In this study, a custom-designed arterial spin labelling RF coil was assessed for its efficiency and safety, and thus is of particular interest to ultra high field MRI scientists working on RF coil design and parallel transmission (pTx). Purpose: RF coils at 7T can be designed to mitigate B1 field inhomogeneity but the most effective solution to RF inhomogeneity is to use parallel transmission. However, such a setup may compromise the safety of the subject through excessive RF heating. In this study, a variant of the pseudo continuous arterial spin labelling (pCASL) [1] sequence was implemented on a Siemens 7T scanner with 8 independent power-supervised Tx channels. A comparison of tagging efficiency and RF heating deposition was made between tagging with a custom built 4-channel Tx neck array versus tagging with a commercial single transmit channel 1Tx/32 Rx volume head coil (Nova Medical, Wilmington, MA). Figure 1 Methods: A 4-channel neck array was constructed using two pairs of loops placed at each side of the neck, held together with a flexible cradle [Fig 1]. For each coil, a pair of accessible variable capacitors was used for fine tuning and matching and the channels were tuned/matched at 297.2MHz and  -20dB. In order to use both the 32ch coil and the transmit labelling array, a special setup [Figs 2 & 3] was necessary to overcome scanner restrictions. A 4-way combiner was used to combine the power of four of the available pTx channels and a switching arrangement used to divert the Tx channel of the 32Ch coil to the combiner [Fig 1]. SEMCAD X (SPEAG, Zurich) software was used for electromagnetic Figure 4 characterisation of the coil in conjunction with a human phantom [3] [Fig 4]. E and B1+ fields per volt per channel were calculated for each of the 4 elements and were then processed with MATLAB for estimation of relative amplitudes and phases for RF shimming around the carotid arteries [Fig 5]. The transmission power limit per channel was set to be the worst case scenario, which is the phase combination that results in the maximum generated SAR [4]. The SAR is calculated over 10g mass cubes, averaged according to the IEEE/IEC 62704-1 standard [Fig 6]. Finally, ASL tagging efficiency was measured experimentally using two flowing tubes with a pCASL prepared FLASH sequence. A spherical phantom was used to load the head coil and a cylindrical phantom used to load the neck coil [Fig 3]. The mean velocity of the flowing water was 12.5 cm/s in a 4mm diameter tube. A flow compensated FLASH acquisition was prescribed with the phase encode direction perpendicular to the flow, and with FOV 200mm, 25% phase encode FOV, slice thickness=5mm, 192x48 matrix, TR=3sec, TE=3.1ms. A 400ms pCASL pulse train preceded each excitation with 0.6ms RF pulses, 1060ms separation, tag gradient Figure 3 6mT/m, mean gradient 0.8mT/m and 10ms post label delay. Tagging was performed with the neck and head coils and the tagging efficiency was measured by taking the complex subtraction of the tag and control images after phase unwrapping using PRELUDE (FSL, Oxford) [5] and after normalising to an M0 image with the tagging pulse voltage set to zero. Results & Discussion: Simulations: In the RF unshimmed condition the anterior/posterior variation in the middle of the ROI is ±0.2μΤ around the mean and B1max/B1min=2.3 whereas in the shimmed condition a ±0.05μΤ variation around the mean was achieved with B1max/B1min=1.2 [Fig 5]. The worst case scenario, as defined in the Methods section, generated a maximum local SAR of 2.87W/kg for 1 Watt per channel input [Fig 6]. Thus, for 1st level control (20W/kg max [6]), the power scale factor is 7 per channel. In the shimmed condition with optimized amplitude and phase variation the max 10g SAR was 2.2W/kg. Phantom Experiments: The effective global SAR of the affected body parts (head, neck & shoulders represented by the spherical and cylindrical phantoms) was calculated from the known power transmitted to the RF coils (and accounting for the 8% duty cycle used in the sequence). During tagging with the 32Ch coil the highest estimated global SAR was 1.98W/kg and the mean ASL tagging efficiency was just over 31%. When the neck coil was used for tagging the highest estimated global and local SAR was 2.5W/kg and 7.3W/kg, respectively, and the tagging efficiency was improved to approximately 72% [Fig 7 & Table 1]. Figure 7 Conclusion: For the same power output, the neck coil provided more than double the tagging efficiency albeit with a 27% SAR Table 1: Tagging Efficiency and SAR [W/kg] increase in comparison to the 32Ch head coil. As long as local SAR Coil Left Tube Right Tube Combined SAR% and power transmission are supervised in total and per channel it is 61% 32Ch 34.1 ± 6.8 % 27.2 ± 4.5 % 31.3 ± 6.9 % preferable to use local tagging for ASL at 7T. Human scanning is 80.2 ± 13 % 59.5 ± 6.3 % 71.9 ± 15 % 78% Neck being conducted to further investigate the benefits from local tagging. References: [1] Tom Okell DPhil Thesis, Oxford University. [2] Mispelter J et al. NMR Probeheads, London 2009. [3] Christ A et al, Phys. Med. Biol. 55, N23, 2010, 2011 [4]Eichfelder G and Gebhardt M, Mag. Res. Med. 1476, 1468-1476, 2011. [5] S. M. Smith et al. NeuroImage 23(S1):20819, 2004. [6] IEC, Medical Electrical Equipment, 60601-2-33, 2010 Cortical thickness map: an automatic quantification of cerebral cortex for in vivo mouse brain MRI 1 D. Ma1,2, M.J. Cardoso1, M.F. Lythgoe2, and S. Ourselin1,3 Centre for Medical Imaging Computing, 2Centre for Advanced Biomedical Imaging, 3Dementia Research Centre, University College London, UK Background The cerebral cortex is the outermost layer of the cerebrum consisting of grey matter. A number of neurodegenerative diseases, such as Alzheimer’s disease, are associated with variations of cortical pathological, especially the global and local thickness of the cortex. [1] Different studies has been conducted to develop cortical thickness model from human brain structural MRI, [1][2] but few research have been done to extract such information from mouse cortex, and correlate it to the pathologies of the corresponding diseases. Objective The objective of this study is to derive an automatic method to construct a map across the cortical region which can represent local cortical thickness based on the automatic parcellated brain structures including cerebral cortex on in vivo mouse brain MRI. Methods We modeled the local cortical thickness as the length of unique non-intersected streamlines connecting the inner and outer boundary of the cortex. By assigning two different “potential values” to the two boundaries (0 and 1000 in our case, without unit), the streamlines can be defined as orthogonal to all the equipotential surface in between the boundaries. This streamline definition of thickness has been shown to have a good correspondence with the physiology of the column structure in the cortex revealed from both functional and histological studies. [1][3] Brain structure extraction: Brain structures are obtained automatically using our previously developed pipeline. [4][5] All the non-cortical cerebral structures are then grouped (including white matter, intra-cortical cortical cerebrospinal fluid and the gray matter in cerebellum). Boundary definition: based on the information from the brain structural labels, the white/grey matter junction, as well as the boundary between the grey matter and supra-cortical cerebrospinal fluid (the pial surface) defined as inner and outer boundary. Two distinct potential values are then assigned to the two boundaries. The structural labels has been separated into left and right hemisphere to prevent the necessity of adding an extra resistance layer in the middle as the method Lerch et al. described [5], and the thickness is calculated separately for each hemisphere. Cortical construction: A gradient field between inner and outer boundary of the cortex is generated by solving the second order partial differential equation (Laplace equation) giving the boundary conditions defined in the previous step. The gradient at each voxel is then normalized and a unit vector field is created in the cortical region. Thickness map integration: The thickness is calculated at each voxel in the cortex region. The length of the streamline is derived by integrating towards both the inner and outer boundary. After they both reach the boundaries, the length of the two parts were added together, resulting the thickness corresponding to the voxel. Faster methods to derive the path length, such as solving with second order Partial Differential Equation will also be implemented to in the later stage. Results and Conclusions The proposed method has been applied and tested on phantom data with structure shape similar to mouse cerebral cortex. Result of thickness map showed reasonable correspondences with the underlining geometry. Data from existing mouse brain MRI atlas [6] will be tested in the next step. Future work might also incorporate tissue classification data which consist the probability map and reduce the partial volume effect. Acknowledgements This work was undertaken at UCL which received a proportion of funding from Faculty of Engineering funding scheme. References [1] Jone, et al. Human Brain Mapping, 2000. [2] Yezzi, et al. IEEE Transactions on Medical Imaging. 2003. [3] Lerch et al. NeuroImage, 2008. [4] D. Ma, et al. British Chapter ISMRM, 2012.[5] D. Ma, et al. MICCAI 2012 Workshop on Multi-Atlas Labeling. [6] Ma, Y., et al. Frontiers in neuroanatomy, 2008. Comparison of the Reproducibility of a GABA-Edited Magnetic Resonance Spectroscopy Technique with and without Macromolecule Suppression Mark Mikkelsen1, C. John Evans1, Petroc Sumner1, Krish D. Singh1 1 Cardiff University Brain Research Imaging Centre (CUBRIC), School of Psychology, Cardiff University, Cardiff, UK Background & Aims: The concentration of the major inhibitory neurotransmitter γ-aminobutyric acid (GABA) can be quantified in vivo using 1 H magnetic resonance spectroscopy (1H-MRS) and has provided insights into pathology and healthy behaviour. However, a problem with the quantification of GABA through standard GABA-MRS techniques is that macromolecules (MM) contaminate the GABA signal. This is caused by MM resonances at 1.7 ppm being coupled to one of the GABA methylene groups (CH2) that resonates at 3.0 ppm. Using a symmetric editingbased suppression technique, however, it is possible to account for MM contamination by suppressing the MM signal that couples to GABA [1]. The aim of this study was to evaluate and compare the reproducibilities of GABA-edited MRS with and without macromolecule suppression. Method: GABA concentration was measured in a single voxel (30 x 30 x 30 mm3) placed in the medial occipital lobe of ten healthy participants (22-39 years, SD = 5.2 years; 7 females) using a 3T GE Signa HDx MRI scanner with an eight-channel receive-only head coil. Four 10-min MEGA-PRESS [2] acquisitions were taken in each participant within a single session. Two of the four scans used a standard MEGA-PRESS sequence (TE = 68 ms; TR = 1800 ms) with 16-ms editing pulses placed at 1.9 ppm (ON scan) and 7.5 ppm (OFF scan), which does not remove MM contamination. The remaining two scans (TE = 80 ms; TR = 1800 ms) used a symmetric MM-suppression method [1] where editing pulses (20 ms) are placed symmetrically about the MM resonance at 1.7 ppm (at 1.9 ppm [ON] and 1.5 ppm [OFF]). The MM resonance peak is thus inverted equally, suppressing the signal in the difference spectrum (ON – OFF). MM-unsuppressed and MM-suppressed scans were interleaved and counterbalanced across participants. Spectra were processed and analysed in MATLAB-based software (GABA Analysis Toolkit [Gannet], gabamrs.blogspot.co.uk). GABA concentration was referenced to internal water, with corrections applied for relaxation times of water and GABA, editing efficiency and MR-visible water concentration. As the standard sequence does not suppress MM, the quantified GABA measure includes an MM contribution and therefore is commonly denoted as GABA+, whereas the suppression method produces a GABA peak corrected for MM contamination (GABAcorr). Results: Mean GABA+ was 1.1 ± 0.08 institutional units (i.u.) and mean GABAcorr was 0.6 ± 0.1 i.u. Similar to previous findings, the GABAcorr concentration was 48.7% relative to GABA+. Within-session coefficients of variation (CV; SD/M) were 4.5% for GABA+ concentration and 8.8% for GABAcorr. These were not significantly different, t(9) = –1.22, p = .25. Between-session CVs were 7.4% and 17.5% for GABA+ and GABAcorr, respectively. Intraclass correlation coefficients (ICCs) were higher for GABAcorr values (.79) than for GABA+ (.71). Conclusions: Symmetric editing-based MM-suppression is shown to be acceptably reproducible. In addition, its comparatively higher ICC may suggest that it is more sensitive to individual variability of MRS-measured GABA. The importance of measuring a “purer” quantification of GABA becomes increasingly important as more research is conducted on direct correlational links between GABA concentration and healthy behaviour. References: [1] Henry et al., Mag. Reson. Med. 45:517-520 (2001). [2] Mescher et al., NMR Biomed. 11:266-272 (1998). Improved Confidence in Diffusion Metrics from ‘Post-Navigator’ Registration of Individual Coronal Signal Average Images in Abdominal Diffusion-Weighted MRI 1 NP Jerome1, JA d’Arcy1, MR Orton1, D-M Koh2, MO Leach1, DJ Collins1 Department of Radiotherapy and Imaging, The Institute of Cancer Research, Cotswold Rd, Sutton, Surrey, UK 2 Department of Radiology, Royal Marsden Hospital, Downs Rd, Sutton, Surrey, UK Background: Respiratory motion commonly confounds abdominal diffusion-weighted (DW) MRI, producing image blurring as a function of the number of b-values and signal averages acquired. Strategies to minimise the impact of respiratory motion can adversely affect scan efficiency (e.g. navigator-triggering) or patient comfort (e.g. breath-hold) without necessarily improving diffusion parameter estimation (1), and rely on potentially invalid assumptions about breathing patterns and consistency. Respiratory motion in the abdomen is predominantly head-foot; acquiring and storing coronal signal averages separately allows for post-acquisition registration of each image, removing the largest source of blurring with no time cost, and potentially the ability to cull aberrant images that may show through-plane respiratory motion or other effects. Figure 1: Mean b=0 images before (left) and after (right) voxel-‐wise shifting Aim: To apply voxel-wise ‘post-navigator’ alignment for IVIM-model DW-MRI images, and examine residuals from model fitting that indicate confidence in calculated diffusion parameters. registration; arrows indicate blurred boundaries. Methods: Healthy volunteers were recruited and imaged on a MAGNETOM 1.5T Avanto (Siemens AG, Healthcare Sector, Erlangen, Germany). Coronal slices covering the kidneys, with parameters: 2D bipolar DW EPI sequence, TR 4000 ms, TE 72 ms, voxel size 1.5 mm2 in-plane, 16 contiguous 5mm slices, 128 x 128 matrix with GRAPPA factor 2, interpolated to 256 x 256, with nine b-values (0, 20, 40, 60, 80, 100, 250, 500, 750 s/mm2) and trace images calculated (NSA 5 or 6, approx. 10 minutes total). Individual trace images for each series and bvalue were registered manually (focusing on the left kidney) to the mean b=0 image by use of pixel-wise shifting. Outliers were defined where the shift magnitude was found to be greater than (mean + 2×s.d.). IVIM diffusion model was fitted for raw images, registered images, and registered images without outliers, using all b-values in a Bayesian approach with in-house software. Calculated diffusion parameters for ROIs of whole kidney, renal cortex, and renal medulla are reported as mean ± s.d., and compared using a paired t-test with significance level 0.05. Figure 2: ladder plots for residual sum-‐of squares from IVIM model fitting. Dashed line is mean (n=10). Results: Manual registration of the individual trace images yields a visually sharper image for the kidney considered (figure 1). Diffusion parameters derived from the left kidney (table 1), show no significant difference in any of the parameter values reported; the only exception is for the pseudo-diffusion constant D* in the renal medulla. The normalised residual sum-ofsquares, a proxy for the confidence of the fitting, is reduced with registration for every subject (figure 2), significantly in both whole kidney and renal medulla, and approaching significance in the cortex (p=0.06). The definition and exclusion of outliers as applied here appears to convey a small further decrease in residuals, significant for the whole kidney ROI, and making the improvement for the renal cortex just short of significant (p=0.051) when compared to nonprocessed images. Discussion: Sufficient signal averaging is used to give robust ROI DWI statistics, and registration/culling of individual images, practiced for DTI in the brain(2), does not change the mean values reported for IVIM parameters. Simple whole-pixel translation alignment as presented here can be used to largely account for the observed head-foot motion and improve image appearance; the scheme is attractively simple, does not interpolate raw data, and offers the lion’s share of blurring reduction whilst being fast enough to automate in the normal workflow. Greater confidence in fitted parameters, indicated by decreased fitting residuals, may offer greater sensitivity of DWI measurements to small changes arising from disease progression or treatment response, and potentially facilitate a move from ROI statistics to interrogation of pixel-maps. Other criteria for identifying and culling outliers from the dataset, not yet explored, may yield further improvements. Acknowledgements: This work was funded by CR-UK grant number C7273. We also acknowledge the support received for the CRUK and EPSRC Cancer Imaging Centre in association with the MRC and Department of Health (England) (grants C1060/A10334 and C16412/A6269) and NHS funding to the NIHR Biomedical Research Centre. We also thank Dr. Thorsten Feiweier at Siemens AG, Healthcare Sector for providing the prototype DWI package. References: (1) Jerome, NP; Orton, M; d'Arcy, J; Collins, DJ; Koh, D-M; Leach, MO, Proc. ISMRM 2012, Abstract 1316. (2) Chang LC, Jones DK, Pierpaoli C., Magnetic Resonance in Medicine, 2005;53, 1088 Automated high-throughput morphometric phenotyping of mouse brains and embryos N.M. Powell1,2, M. Modat1, M.J. Cardoso1, D. Ma1,2, H. Holmes2, F. Norris2, M.F. Lythgoe2, S. Ourselin1. 1 Centre for Medical Image Computing, University College London, UK, 2Centre for Advanced Biomedical Imaging, UCL Background Microscopic MRI (with resolution <150µm) has in recent years enabled the transfer of established statistical morphometric analysis techniques used in human studies, such as voxel- and tensor-based morphometry (VBM, TBM), to preclinical studies involving rodents [5]. VBM and TBM may be used to naively detect local structural changes in tissue based upon MRI signal – that is, without prior specification of a region of interest, making them ideal for investigative studies of gene function and disease. Robust, automatic phenotyping tools employing these techniques to ease the labour and time burden of manual segmentation in preclincal studies are not presently available. Such tools will be essential for the efforts of the International Knockout Mouse Consortium in identifying the morphometric influence of some 25,000 individual mouse genes [5,7]. Furthermore, validation of mouse models of disease, particularly via their correspondence with known morphometric changes in humans – in the adult mouse, developing brain, and embryos – is essential prior to observation of disease progression and drug treatment. VBM employs prior maps of grey and white matter structure for detecting tissue density changes in the brain associated with degenerative diseases such as Alzheimer’s. TBM is used to detect volumetric changes in physical structures [1,2]. Both have been applied to preclinical rodent scans [5,8,10], but TBM has until now not been applied to embryos. Objectives We aim to develop a robust, high-throughput, fully automatic morphometric pipeline for µMRI images of the mouse brain (with VBM and TBM) and embryo (TBM), which may be verified with histological investigation of regions of the most significant difference between groups, and which is compatible with automated segmentation propagation. Methods We obtain in- and ex-vivo µMRI images using optimised scanning protocols for mouse brain and embryo phenotyping, on a Varian 9.4T scanner [6]. We typically simultaneously scan up to 3 in-skull brains or 9 embryos ex-vivo, using custom holders (e.g. Fig. 1). The image processing pipeline then follows the following fully automatic steps [1,2,5]: Extract multiple subjects into individual images using an iterative expectation maximisation approach to separate intensities [3], automatically discard extraneous objects, and crop. II. Align subjects into standard RAS orientation using their principal axes, and symmetry. III. Brains: Mask using a STAPLE scheme [11] and existing atlas labels. IV. Perform nonuniformity correction, and standardise intensities. V. VBM: Initialise tissue probability maps (TPMs) of GM, WM and CSF using an expectation maximisation algorithm [3] with tissue class priors. VI. Produce an average atlas of the study population and deformation maps via iterative groupwise registration, using the open-source NiftyReg [9] VII. VBM: Resample TPMs to the final average atlas, modulate, and smooth [1]. VIII. TBM: Transform deformation maps prior to statistical analysis, taking the logs of the determinants of the Jacobian matrices of each deformation field, and smooth. IX. Perform investigative statistics on the deformation maps (TBM) and TPMs (VBM) using Fig. 1: GLDC embryos ANCOVA and the General Linear Model. We produce volumetric statistical parametric maps prior to extraction (SPMs) with FDR-corrected t- and F-statistics to show significant differences between groups’ morphometry. I. An automatic segmentation propagation pipeline [4], using the resulting average atlas and deformation maps, enables the quantification of volumetric changes in specific parcellated structures of both brains and embryos, using prior parcellated mouse embryo [6] and brain atlases. Results This pipeline represents the first step toward a freely-available automatic image processing platform for the structural analysis of mice, from scanner to statistics. Currently, there is no such integrated software available. We have applied our software to both mouse embryos and adult brains in- and ex-vivo. This platform enables the end user to perform phenotyping assessment without the need to move between packages or develop in-house software. We demonstrate our pipeline on the following mouse strains: GLDC (a model of spina bifida) and TBx1 (a gene implicated in embryonic development) embryos; Tc-1 (a model of Down Syndrome); NK1 (a model of ADHD); PML (a gene encoding a tumoursuppressing protein); and TG4510 (a model of Alzheimer’s). Conclusions We have established an optimised scanning and image processing pipeline for the high-throughput phenotyping and morphometric study of rodent embryos and brains, enabling the verification of mouse models, the longitudinal observation of disease progression and drug therapy, as well as the detection of unforeseen changes in tissue structure caused by genetic abnormalities or behavioural adaptation. This pipeline is ideally suited to both ex-vivo and optimised in-vivo scans of mouse brains, enabling automated statistical analysis of high-resolution longitudinal morphometric studies. This represents a step change in mouse phenotyping and the opportunity to meet the need of the International Knockout Mouse Consortium. Acknowledgements Supported by UCL & MRC studentships; the NC3Rs; Eli Lilly; the British Heart Foundation; & EPSRC. References [1] Ashburner, J., et al. (2000). NeuroImage, 11(6 Pt 1), 805–21. [2] Ashburner, J., et al. (2000). NeuroImage, 11(5), S465. [3] Cardoso, M.J., et al. (2011). NeuroImage, 56(3), 1386–97. [4] Cardoso, M.J., et al. (2011). MICCAI MLSF. [5] Cleary, J.O. et al. (2011). Doctoral thesis. [6] Cleary, J.O., et al. (2011). NeuroImage, 54(2), 769–78. [7] Collins, F.S., et al. (2007). Cell, 128(1), 9–13. [8] Lerch, J.P., et al. NeuroImage, 54(3), 2086–95. [9] Modat, M., et al. (2010). Comp. methods and programs in biomed., 98(3), 278–84. [10] Sawiak, S.J., et al. Neurobiology of disease, 33(1), 20–7. [11] Da M., et al. (Proceedings) MICCAI 2012 Grand Challenge and Workshop on Multi-Atlas Labeling. GlucoCEST for the detection of human xenograft glioblastoma at early stage 1 1 1 1 2 1 F. Torrealdea1, M. Rega , A. Richard-Loendt , S. Brandner , D.L. Thomas , S. Walker-Samuel , X. Golay 1. Institute of Neurology, 2. Centre for Advanced Biomedical Imaging, University College London, London, UK Introduction The high glucose uptake in tumours (Warburg effect) is exploited by FDG-PET to detect tumours and indicates maximal rates of cell proliferation1. GlucoCEST is a newly developed MRI technique with the ability to detect glucose in tissues, based on the chemical interaction of glucose hydroxyl protons and water. Previously reported, glucoCEST and FDG measurements provide equivalent information in various flank tumour models2. Deregulated glial metabolism of primary brain tumours is likely to alter the dynamics of the glycolytic pathway causing 3 changes in the homeostatic levels of glucose . GlucoCEST could provide contrast between tumours and the surrounding 4,5 brain tissue based on differences in glucose levels, unlike FDG-PET which reports specifically on uptake . This work explores the feasibility of using glucoCEST as a tool for early detection of primary brain tumours. Methods Human glioblastoma cells were injected intracranially in immune suppressed (NON-SKID) mice and allowed 200 days to grow. Mice were fasted for 24 hours in order to reduce and stabilize blood glucose levels, anaesthetized with 1.3% isoflurane and cannulated via the intra peritoneal route for glucose administration (1g/Kg) while in the scanner. GlucoCEST was acquired using a turbo-flash sequence with a saturation train prior the readout of 80 Gaussian pulses at 1µT. GlucoCEST enhancement (GCE) 2 was calculated as the signal change between pre- and post- glucose administration. High-resolution spin echo (SE) anatomical scans and histological analysis of human-specific cells (Vimentin) were performed for comparison with glucoCEST images. Results and Discussion A significant signal increase (p<0.03) was observed in regions affected by tumour, detected from 12 minutes after injection (Fig.1). The averaged GCE image over the first 20 minutes shows a well-demarcated tumoural area. Comparison of the GCE with anatomical images and histology suggests that glucoCEST can identify tumours earlier than conventional MRI (Fig.2). A possible explanation is that at early stages of cancer, while brain structures are still not disrupted and the T2 relaxation times are unaffected, tumours cells already proliferate at higher rates. Pixel by pixel analysis of GCE at day220 versus anatomical image at day235 shows a weak correlation (R2=0.13) but no correlation (R2=0.027) at day220. Conclusion This study shows that glucoCEST is sensitive enough to distinguish between cancerous and healthy tissue in the brain. Due to the particular source of glucoCEST contrast (glucose uptake rather than relaxation times), the technique can depict cancer-affected areas before the appearance of microstructural changes. Furthermore, the technique allows dynamic assessment of tumour metabolism, which might be useful for the characterization of tumour malignancy. Further work will be needed to investigate the importance of glucose uptake dynamics for the characterisation of glial tumour grade. References [1] Marin-Valencia et al. Cell Metab. 2012 Jun;15(6):827-37. [2] Walker-Samuel et al. Nature Medicine. 2012 In press. [3] Torrealdea et al, ISMRM 2012. [4] W. Y. Chan et al, MRM 2012. [5] Basu et al. Neuroimaging Clin N Am. 2009 Nov;19(4):625-46. 2 GCE% 1.5 T2w-SE / Day 220 GlucoCEST / Day 220 T2w-SE / Day 235 Histology / Day 235 1 0.5 0 Tumour region Healthy region -0.5 0 30 60 90 120 [minutes] Figure 2 Time evolution of the mean GCE across two different regions (tumour in red, healthy tissue in blue). A significantly higher signal increase is observed in the tumour region followed by glucose administration at t=0 minutes. Figure 1 Comparison between glucoCEST, spin echo and histology. At day 220 post tumour injection, the SE image does not show the full spread of the tumour, while glucoCEST already displays the features that will be detectable 15 days later by T2w SE. The overlay of MR and histology slice shows tumour cells highlighted in brown by a human specific stain (Vimentin), whereas the host brain is shown in blue (Haematoxylin counterstain). THE IMPACT OF GROUP-WISE DIFFUSION TENSOR REGISTRATION ON TRACT-BASED SPATIAL STATISTICAL ANALYSIS OF WHITE MATTER MICROSTRUCTURE IN NEONATAL ENCEPHALOPATHY Authors 1 2 3 2 1 Lally PJ , Price DL , Zhang H , Cady EB , Thayyil S Institution(s) 1 2 Academic Neonatology, University College London, London; Medical Physics & Bioengineering, 3 University College London, London; Centre for Medical Imaging and Computing, University College London, London Introduction White matter (WM) fractional anisotropy (FA) analysed by tract-based spatial statistics (TBSS) correlates with adverse neurological outcome following neonatal encephalopathy (NE) [1-4]. Recently, between-subject registration of FA maps to a group-wise atlas has been shown to improve sensitivity and specificity of TBSS analysis in neurodegeneration [5] compared to registration to the most representative subject within the group, as is typically used. Registration of individual diffusion tensor maps to a group-wise atlas may similarly improve the robustness of the TBSS pipeline, and hence the sensitivity to pathological FA change in various NE-associated comorbidities. We assess the impact of a group-wise atlas together with tensor-based registration on TBSS analysis in NE. Methods Consecutive infants with NE admitted to Calicut Medical College over a 6-month period were recruited after parental consent. Conventional MRI and diffusion tensor imaging (DTI, 21 directions, b = 0 2 &1000 s/mm ) were performed at 1.5 Tesla (Siemens Avanto, Erlangen, Germany). TBSS analysis had previously been performed to investigate correlation with comorbidities via non-linear registration of each subject FA map to the most representative target [4] using the FMRIB Software Library (FSL) [6-7]. This process was repeated in three of these comorbidities, instead using DTI-TK [8] for non-linear registration of the diffusion tensors of each subject to an iteratively optimised group mean template prior to statistical analysis (see Figure). Skeletonised mean FA maps (green in Figure) were thresholded to ensure equal volumes. For each b) comorbidity, the t-scores in voxels showing group-wise a) difference in FA (p<0.05, uncorrected for multiple tests) Figure Mean FA across all subjects after were used to assess the relative sensitivities of the each registration method: a) most representative subject b) group-wise processing techniques. Results Thirty-one subjects had usable TBSS data and were grouped according to the presence of each comorbidity. The group-wise atlas generated by tensor registration was distinctly crisper than when using the most representative subject for registration (see Figure), suggesting improved alignment of individual WM tracts. The number of voxels reaching the p<0.05 threshold was reduced in each case when using this group target, although not to a significant level. This may be explained by the th th artificially inflated power associated with FA-to-FA registration [9]. Median, 75 and 95 percentile tscores across all significant voxels also showed no significant difference between the two methods. Conclusions Non-linear tensor registration to a group-wise template shows no significant sensitivity change in comparison to the use of the most representative subject. A group-wise template may still be preferred to minimise bias in TBSS analysis, although no considerable morphological differences were observed between subjects in this cohort. References [1] Tusor N et al. 2012 Pediatr Res 72(1):63-9 [2] Thayyil S et al. 2010 Pediatr Res 125(2):382-95 [3] Thayyil S et al. 2011 Neonatal Society Autumn Meeting, London [4] Price DL et al. 2012 Proceedings of Eur Soc Magn Reson Med Biol 29th Annual Meeting, Lisbon, p114 [5] Keihaninejad S et al. PLoS One 7(11):e45996 [6]Smith S et al. 2004 NeuroImage 23(S1):208-19 [7] Smith S et al. 2006 Neuroimage 31(4):1487-505 [8] Zhang H et al. 2006 Med Image Anal 10:764-85 [9] Tustison NJ et al. 2012 Human Brain Mapp Nov 14 Multi-Slice Look-Locker FAIR for Hepatic Arterial Spin Labelling R. Ramasawmy1,2*, A. Campbell-Washburn1*, S.P. Johnson2, J.A. Wells1, R.B. Pedley2, S. Walker-Samuel1†, M.F. Lythgoe1† 1 2 UCL Centre for Advanced Biomedical Imaging, London, UK UCL Cancer Institute, London, UK Purpose: Arterial spin labelling (ASL) is used in the brain [1], heart [2] and kidney [3] to measure perfusion but has not yet found extensive utility in the liver, due to its dual vascular supply and susceptibility to respiratory motion. Non-invasive liver perfusion measurements could monitor hepatic disease progression and drug efficacy in pre-clinical models of cirrhosis [4] and tumour metastasis [5]. Previous work demonstrated single-slice Look-Locker Flow-Sensitive Alternating Inversion Recovery (FAIR) hepatic ASL measurements [6]; however a multi-slice perfusion sequence would increase efficiency of whole liver coverage when imaging multiple metastases and gross liver dysfunction. In this study we demonstrate the use of a multi-slice Look-Locker FAIR ASL and compare it to equivalent singleslice perfusion data. Methods: ASL acquisition: Single slice perfusion measurements were obtained using a respiratory-triggered inversion, segmented FAIR Look-Locker ASL sequence with a spoiled gradient-echo readout [6]. The multi-slice sequence was adapted from the single-slice technique with additional segmented acquisition pulses for each slice within the Look-Locker train [7]. Multi-slice sequence parameters were: FOV 30 x 30 mm2; matrix size 128 x 128; 3x1 mm slices with 0.2 mm gap, TE 1.18 ms; TI 110 ms; TRRF 2.3 ms; αLL=8°; TRI 13 s; 50 inversion recovery readouts, 4 lines per segmented acquisition, 15 minute acquisition time. For both single- and multi-slice acquisitions, a localised 6 mm slice selective inversion centred on the middle slice was followed by a global inversion. Scans were performed on a 9.4T Agilent VNMRS 20 cm horizontal-bore system, using a 39 mm birdcage coil. Inversions were triggered at the end of the inspiration phase using respiratory gating apparatus (SA Instruments, US). In vivo measurements: Three mice were anaesthetised using 1.5% isoflurane in 100% O 2 and positioned in the centre of the magnet. Core body temperature was monitored and maintained using a warm air blower. Respiratory-gated fast spin echo images were used to define suitable axial imaging slices within the liver. Post-processing: Perfusion maps were calculated using the model as described by Belle et al [2]. A blood-tissue partition coefficient of 0.95 ml/g was taken from 85 Kr gas clearance measurements [8]. The liver capillary blood T1 was assumed to be 1900 ms, from previous T1 measurements of the ventricular blood pool in the mouse heart [9]. Perfusion to the liver is assumed to be delivered from both the arterial and venous systems. Results: Fig. 1 shows three slices through a murine liver: for each column there is an anatomical T2-weighted image (Row A) of the liver above its associated single slice (Row B) and multi-slice perfusion map (Row C). On the anatomical images, the stomach and blood vessels appear hypo-intense compared to liver tissue. The major vessels can be visualised in Fig. 1B & 1C due to a large but non-physiological perfusion signal. The multi-slice perfusion maps (mean perfusion pMS = 2.1 ± 0.8 ml/g min) compares well with the single slice (mean perfusion pSS = 1.9 ± 0.7 ml/g min) though BlandAltman analysis suggests a slight overestimation in the multi-slice with a mean difference of 0.25 ml/g min. Our multi-slice liver perfusion values are comparable with 85Kr gas clearance measurements [8]. A B ml/ g min Discussion & Conclusions: Arterial spin labelling has been principally used for measuring brain perfusion [2], with more recent application to cardiac [3] and renal [4] imaging. We have previously shown the feasibility of localised liver perfusion measurements using FAIR-ASL [6], an application that has not C been extensively reported in the literature, and here demonstrate an improvement to this technique with a multi-slice adaptation. For these data the multi-slice sequence offers a threefold increase in time efficiency for the same liver coverage as the sequence takes the same amount of time as a single slice acquisition (less than 15 minutes); the sequence could easily be adapted to cover more slices. The slight perfusion overestimation measured could be corrected with a more appropriate quantification method which Figure 1: Three T2-weighted, fast spin echo images of a liver at the different accounts for inflowing blood magnetisation [7]. The perfusion maps generated slice positions with the liver ROI outlined (Row A). Corresponding single-slice are from a mixture of both the arterial and portal systems; a pseudoperfusion maps (Row B) and multi-slice perfusion maps (Row C). Visual continuous ASL method could be implemented to evaluate their respective inspection indicates good correlation between the two techniques; high flow can contributions. Using this sequence, we aim to investigate perfusion changes be seen at major blood vessels such as the portal vein (long arrow) and inferior in colorectal cancer metastasis induced by novel anti-cancer therapies. vena cava (short arrow). Furthermore, brain and kidney FAIR ASL is commonplace in clinical scanners, and given the non-invasive nature of the technique, we anticipate that translating hepatic multi-slice Look-Locker FAIR ASL into a clinical setting would be straightforward. *Joint First Authors †Joint Senior Authors Acknowledgements: This work was supported by an MRC Capacity Building Studentship, the British Heart Foundation, King’s College London and UCL Comprehensive Cancer Imaging Centre CR-UK & EPSRC, in association with the DoH (England). References: [1] Golay X, et al. Top Magn Reson Imaging. 2004; 15:10-27. [2] Belle V, et al. J Magn Reson Imaging 1998;8;1249-1245. [3] Karger N, et al. Magnetic Resonance Imaging. 2000; 18:641-647. [4] Van Beers B, et al. AJR. 2001;176:667-673. [5] de Bazelaire C, et al. Clin Cancer Res. 2008; 14:5548-5554. [6] Ramasawmy R. et al. Proc Intl Soc Reson Med 2012 20:2900. [7] Campbell-Washburn A, et al Magn Reson Med 2012 (in press). [8] Rice G et al, J Pharmacol Methods. 1989 Jul;21(4):287-97. [9] Campbell-Washburn A, et al. Magn Reson Med. 2012; doi. 10.1002/mrm.24243. Measuring myocardial velocities with high resolution using retrogated spiral phase velocity mapping Authors: Robin Simpson, Jennifer Keegan and David Firmin – Imperial College London and NHLI BRU (Royal Brompton Hospital) Background: Measuring regional myocardial motion potentially allows a better understanding of the health of the heart than global parameters of function such as ejection fraction. One of the available methods for measuring regional myocardial motion with MR is phase velocity mapping (PVM). Current high resolution PVM sequences with high resolution use Cartesian k-space coverage in combination with respiratory navigator gating, leading to long acquisition times [1,2]. Also, prospective ECG-gating means that the analysis of the full cardiac cycle has not previously been possible. In this study, a high temporal and spatial resolution PVM technique using efficient spiral k-space coverage and retrospective ECG-gating is presented which allows detailed analysis of the entire cardiac cycle in ten healthy volunteers. Methods: The sequence covers k-space with 13 spiral interleaves (12ms duration, TR 21ms). Navigator-gated reference and 3directional velocity-encoded data (15cm/s in-plane, 25cm/s through-plane) are acquired in consecutive cardiac cycles following a dummy cycle (nominal duration 53 cardiac cycles). Acquired spatial resolution is 1.4x1.4x8mm (reconstructed 0.7x0.7x8mm). Retrospective gating allows full coverage of the cardiac cycle with 60 phases per RR-interval. Basal, mid and apical short-axis slices were acquired in 10 healthy volunteers on two occasions on a Siemens Skyra 3T scanner. Radial, circumferential and longitudinal peak and time-to-peak (TTP) velocities were measured. Reproducibility of peak and TTP velocities was also assessed using Bland Altman analysis. Figure 1: Example basal curves for longitudinal (a), radial (b) and circumferential (c) directions. End systole is marked by a vertical dotted line on the radial curve. The main peaks (systole (S), early diastole (D) and atrial systole (AS) for longitudinal and radial velocities and the three circumferential peaks (C1, C2 and C3)) are marked. Results: The high temporal resolution allowed consistent visualisation of fine features of motion, while high spatial resolution allowed the detection of statistically significant regional and transmural differences in motion. Figure 1 shows example data and highlights the main velocity peaks throughout the cardiac cycle (of particular note is the late diastolic peak which has not previously been seen with prospectively gated PVM studies), while Figure 2 shows the mean peak and TTP velocity values +/- standard deviation for those peaks. TTP values are shown normalized to a fixed systolic (350ms) and diastolic (650ms) length. Peak velocity values similarly show small standard deviations. Figure 3 shows basal, mid and apical short-axis colour-maps displaying regional velocities against time after the R-wave, averaged over the 10 volunteers. The reproducibility of all parameters was good, with no significant differences between the two scans. Figure 2: Mean+/-SD values of peak and TTP velocities for the peaks shown in Figure 1. Peak and TTP values show low standard deviations throughout the cardiac cycle. C3 is not present in the mid slice. Figure 3: Basal (top), mid (middle) and apical (bottom) short-axis colour-maps displaying regional longitudinal (left), radial (middle) and circumferential (right) velocities (from anterior wall through to lateral, inferior and septal walls on the vertical axis) against time after the R-wave (horizontal axis), averaged over the ten healthy volunteers. Complicated motion patterns can be rapidly visualized. Conclusion: The use of spiral imaging has allowed the acquisition of high resolution PVM images in a relatively short acquisition time (95 ± 16 cardiac cycles per slice), while retrospective cardiac gating has enabled the analysis of the entire cardiac cycle including late diastole (atrial systole). The newly introduced colour plots allow easy interpretation of complicated regional motion patterns and make use of the high spatial resolution acquired. Future work will include implementing parallel imaging to further speed up the acquisition. References: [1]Jung,2006,JMRI;[2]Delfino,2006,JMRI;[3]Leong,2010,JACC Comparison of Bayesian and Linear Regression-based Partial Volume Correction in Single Time Point ASL 1 Ruth Oliver1, Michael Chappell2,3, David Thomas1 and Xavier Golay1 Institute of Neurology, University College London 2Institute of Biomedical Engineering, University of Oxford, 3FMRIB Centre, University of Oxford Introduction: ASL is usually acquired at low resolution, resulting in cerebral blood flow (CBF) measurements that are significantly affected by partial volume (PV) effects. Recent PV correction efforts have focused on a local linear regression (LR) within a kernel to separate voxel signal into GM and WM contributions for single-TI data [1][2]. However, this strategy introduces a spatial smoothing dependent on kernel size. More recently a Bayesian approach has been proposed that exploits the difference in white matter (WM) and grey matter (GM) kinetics and employs an adaptive spatial smoothing by combining kinetic information with PV constraints [3]. However, this latter method has not been applied to the more common single-TI data.This work investigates whether the Bayesian approach offers an advantage over the LR method on single-TI including the extension of the LR method to a 3D kernel. Methods: Assuming no contribution from CSF in difference image, the signal can be modelled as: ∆M = ∆MGMPVGM + ∆MWMPVWM where PVGM and PVWM are the partial volume estimates (PVE) for GM and WM obtained from a structural segmentation from a high resolution T1-weighted scan. The separate tissue contributions can be estimated using LR by making the assumption that they remain constant over the kernel area. This introduces an inherent spatial smoothing based on the kernel size but that also varies with PVE. Previous studies using LR methods have focused on 2D regression kernels, which impose a minimum kernel size due to the need to obtain sufficient equations to solve the LR. Recently it has been shown that a 3D kernel of size 3x3x3 is viable and produces less smoothing than its 2D equivalent [4]. On the other hand, the more recently proposed Bayesian method starts from the same equation but exploits the differing kinetics from the contributions of GM and WM to the difference image, as well as employing a spatial prior to exploit spatial homogeneity in the signal. In single TI data kinetic information is not available; however, spatial priors can still be employed. These would operate like the spatial kernel in the LR method, except the effective amount of spatial smoothing with this method is determined automatically from the data and can adaptively vary across the brain. Data were analysed according to the standard model that does not account for PV effects [6], using the LR method (with kernel sizes in 2D and 3D) and using the spatial prior method using the FSL toolbox (www.fmrib.ox.ac.uk/fsl) implementation [7][8]. Experimental data: Resting state ASL data of 6 healthy controls was acquired using a pulsed Q2TIPS labelling scheme with background suppression [5]. A 3D-GRASE read out was employed with total inflow time = 2s, bolus length = 0.8s, matrix size = 64 x 64 x 20, image resolution = 3.75 x 3.75 x 3.8 mm3. High resolution MPRAGE images were acquired in the same session, matrix size = 208 x 256 x 256, resolution = 1.0 x 1.0 x 1.0 mm3. All images were acquired on 3T Siemens Trio using a 32-channel head coil. The T1-weighted images were segmented using SPM8 into PV Fig1 GM CBF maps (ml/100g/min) fractions for GM, WM and CSF. These were transformed to ASL space via linear registration and resampling to ASL resolution using FSL FLIRT and applywarp as in [3]. To estimate the amount of smoothing each of the methods introduced, the spatial gradient was calculated at each voxel for the PV corrected GM CBF images in all 3 spatial dimensions, the magnitude of this vector was taken as a measure of detail. Regions of interest (ROI) were defined according to the PV fraction in a similar manner to [3], allowing quantitative comparisons to be made of the mean GM CBF and spatial gradient across the PV range for the two methods. The ranges were 10-20, 20-30…90-100%, with the expectation that PV corrected GM CBF would not correlate with PV fraction, but that the mean spatial gradient would reduce for LR methods whilst increasing for spatial prior method as PV estimate increases due to the adaptive nature of the method. Results: Figure 1 shows the middle slice of a subject for each of the analyses. The increased smoothing by LR is readily apparent, although both LR and the Bayesian PV methods perform well for correcting PV, increasing CBF by a factor of 1.6-1.7 as previously found for multi-TI data [3]. (see Table 1). Figure 2 shows the variation in mean CBF with increasing PVE for the same subject. Across all six subjects, there was a distinct trend for each method; LR estimates were usually higher than the Bayesian methods at large PVE, whilst the converse was true for smaller PVE. It is not clear whether LR overestimates at high PVE, or if Bayesian methods underestimate at low PVE. In figure 3 we see a comparison of the mean spatial gradient at each PVE for the Bayesian method, as well as the LR method for a 3D 3x3x3 and a 2D 5x5 kernel. Both LR kernels contain a similar number of voxels and therefore information, yet we see increased retained detail by the 3D kernel as it is drawn from a smaller region of the image. However, both LR kernels are substantially out performed by the Bayesian method, which produced a mean spatial gradient which was a factor of 1.3 and 1.4 greater than the 3D and 2D kernels respectively. Discussion: Adaptive spatial methods appear to be better at retaining detail and reducing smoothing across the whole spectrum of PVE. These advantages must be weighed against the increased complexity of the method. LR is currently faster to compute and provides a comparable average GM CBF to spatial methods, although requires a user selected kernel. The Bayesian PV method adaptively chooses the degree of smoothing, as was evident in the results. This work indicates that both methods are readily applicable to both single and multi-TI data, although currently the LR method when applied using a 3D kernel is best suited to isotropic resolution to avoid greater smoothing in the slice direction. Subject Standard Bayesian LR 1 39.32 61.6 63.6 2 38.8 66.82 71.64 3 33.02 56.75 55.9 4 34.81 55.52 63.49 5 34.25 57.86 63.38 6 30.61 54.00 53.54 Table 1 Mean CBF ml/100g/min for all subjects for each method of CBF calculation. 3D 3x3x3 kernel used for LR. Fig2 Subject 1: CBF ml/100g/min v GM PVE%. LR=red, Bayesian=blue, Std=black Subject Bayesian LR 3x3x3 LR 5x5 1 21.47 18.84 16.93 2 20.23 16.47 15.99 3 18.93 14.01 13.71 4 16.05 13.86 13.20 5 21.64 17.24 16.49 6 27.9 15.09 13.94 Table 2 Mean spatial gradient for all subjects for Bayesian and 2 LR kernel size analyses. Fig3 Subject 1: Spatial gradient v GM PVE%. Bayesian=blue, LR 3D=red, LR 2D=black. References:1.Asllani, I., et al., MRM, 2008. 60:1362-1371. 2. Liang, X, et al., MRM, 2012. 3.Chappell M.A.., et al., MRM, 2011. 65:1173-1183 4.Oliver, RA., et al Proc ESMRMB 2012. 5.Gunther M., et al., MRM, 2005. 54:491-498 6.Buxton, R.B., et al., MRM, 1998. 40:383-396 7.Chappell M.A., et al, IEEE Trans. Sig. Proc. 2009 57(1):223-236 8.Groves, A.R., et al., NeuroImage, 2009 45(3) 795-80 High polarization of nuclear spins mediated by nanoparticles at millikelvin temperatures a b a a a David T Peat , David G Gadian , Kelvin S K Goh Anthony J Horsewill , and John Owers-Bradley a b School of Physics & Astronomy, University of Nottingham, Nottingham NG7 2RD,UK; UCL Institute of Child Health, 30 Guilford Street, London WC1N 1EH, UK, Introduction We report on a novel strategy for generating high levels of nuclear spin polarization. The central notion is that as the temperature is reduced and the magnetic field is increased, the equilibrium nuclear polarization will increase, according to the Boltzmann distribution. For example, at 7 millikelvin (mK) and a field of 16T (conditions achievable with 13 commercially-available technology), the equilibrium polarization of C nuclei would be approximately 50%, i.e. ~200,000fold greater than at 310K and 3T, which are typical conditions for clinical MRS. The main problem with this so-called brute-force approach is that it may take an excessively long time (possibly many years) for the polarization to approach thermal equilibrium at very low temperatures. Here we describe investigations that address this problem. Our goal was to 13 15 13 15 achieve high polarizations of nuclei such as C and N with a view to using pre-polarized C- or N-labelled agents to probe tissue metabolism in vivo. We explored the possibility that selected nanoparticles (including metallic nanoparticles) might act as low temperature relaxation agents. 13 Methods Samples were prepared by mixing nanoparticles with 50/50 water/glycerol solutions containing 2 molar 1- Clabelled sodium acetate and 1 molar sodium phosphate. The volume ratio was one part nanoparticle to 4 or 8 parts of solution, and the resulting mixtures had a wet sandy consistency. Copper (size 25nm), silver (size 20-30nm), aluminium (size 18nm), and graphene (size 11-15nm) nanoparticles were obtained from SkySpring Nanomaterials Inc; platinum (size <50nm) and cupric oxide (size <50nm) nanoparticles were obtained from Sigma Aldrich. Experiments were carried out using a spectrometer that operates at any chosen field up to 15T, and forms part of a dilution refrigerator-cooled system that yields sample temperatures as low as 10mK. 1 Results H T1 measurements at 2.45T revealed that copper, cupric oxide and platinum nanoparticles are highly effective relaxation enhancers at millikelvin temperatures, with copper showing the greatest enhancement. However, aluminium and silver nanoparticles 13 were ineffective. C data were obtained at 9.74T using a volume ratio of 1 part copper nanoparticles to 8 parts solution (see Fig. 1). 13 The estimated C T1/2 value (time to reach 50% of equilibrium polarization) at 19mK was about 40 hours, and was only 3 times as 13 long as the value at 770mK. For comparison, the C T1/2 value measured in the presence of aluminium nanoparticles was at least one year, which gives some indication of the degree of enhancement conferred by the copper nanoparticles. Further experiments were carried out at 14T and 1mK using a 1:4 volume ratio of copper 13 nanoparticles to solution. Under these conditions the C polarization reached 6% after 24 hours, and the T1/2 for growth towards the equilibrium polarization of 23% was estimated to be about 60 hours. 13 Figure 1. C signal x temperature as a function of time. The end-point (normalized to 1.0) should be the same for all timecourses. Discussion At this stage, much remains to be learned about the underlying relaxation mechanism(s). In the presence of the platinum nanoparticles, the proton relaxation times varied inversely with temperature; this accords with the Korringa law for nuclear relaxation in metals, and suggests involvement of the conduction electrons in the relaxation process. However, some of our other findings, including the effectiveness of cupric oxide nanoparticles and the ineffectiveness of aluminium nanoparticles, suggest a role for additional or alternative mechanisms. Interestingly, it has been shown that many nanoparticulate materials (including copper and cupric oxide) display magnetic properties that are not seen in the corresponding materials in bulk form. We suspect that these magnetic properties could be a major factor in the relaxation enhancement that we report on here. Magnetic impurities (including small amounts of iron) may also play a role. Conclusions While further experiments are required in order to establish mechanisms, it is evident that this methodology will enable us to generate and store large-scale quantities of highly polarized materials. A wide variety of applications are envisaged, including investigations of tissue metabolism following dissolution and subsequent administration of pre13 15 polarized C or N-labelled agents. For human studies, an additional feature is that the nanoparticles should be easily removable on dissolution. With further developments, including the use of polarization transfer techniques and simultaneous polarization of many samples, outputs of ten or more samples per day should be feasible. . Acknowledgements We thank Bruker UK for their support. Intracranial (icEEG)-fMRI: mapping brain networks associated with alpha and beta in sensorimotor cortex. Authors and affiliations: 1 1,3 1,2 1 Suejen Perani , Serge Vulliemoz , Roman Rodionov , Louis Lemieux , David Carmichael 1 1 2 UCL Institute of Neurology, London, United Kingdom, National Hospital for Neurology and Neurosurgery, London, United 3 Kingdom, University of Geneva, Geneva, Background and aims: Alpha and beta oscillations have been described as EEG features seen in the frontal-parietal cortex at rest. However, the anatomical and functional relationship between them is still not clear. Whether alpha and beta activity originate in spatially independent regions across the sensorimotor cortex and what their roles are in mapping different brain networks remain unclear. Simultaneous icEEG-fMRI has been implemented as a new technique uniquely capable of overcoming some of the spatio-temporal limitations of EEG and fMRI. It provides the opportunity to study local alpha and beta oscillations locally and their haemodynamic correlates across the entire brain. Methods: One patient undergoing presurgical evaluation with icEEG (1x57 grid) over the left sensorimotor cortex had MRI recording using a 1.5T scanner, head RF-coil, low SAR (≤0.1 W/kg, head average) sequence. T1-volume MRI (TR=3s, TE=40ms, flip angle=90◦), two sessions of 10 minutes (EPI) and one of 5 minutes (5 repeated 30s blocks of finger tapping finger-to-thumb movements left vs. right hands) T2* weighted EPI sequences (TR=3s, TE=79ms, 38slices, 200 volumes, voxel size 3x3x3mm) were acquired during simultaneous acquisition of icEEG data. After application of standard artefact corrections to EEG, the signals from the 48 contacts of interest were transformed into a time-frequency space at frequencies of 8, 9, 10…30 Hz. These were averaged into frequency bands (alpha1, alpha2, beta1, beta2, beta3) showing a similar distribution of the power across time and space. For each band, the first principal component (PCA) was calculated to identify the most representative distribution of the power across time and space. These signals were then convolved with a standard canonical HRF and after standard fMRI pre-processing; they were entered as regressors into a General Linear Model. F-contrasts across alpha sub-bands and beta sub-bands and across both were performed to map the hemodynamic correlates of these EEG features. Results: The power distributions resulting from the first PCs showed that most of alpha1 and alpha2 were from S1 (max at channel 31-finger tapping; channel 48-rest), whilst beta bands were predominantly recorded from M1 (channel 14-finger tapping; channel 13-rest). F-tests showed that alpha and beta bands correlate respectively with right M1 and bilateral S1 during finger tapping, marking the motor network and showing a lateralized correlation with the recording site. During rest, in addition to the sensory-motor network, BOLD signal were found in the following regions to be correlated with alpha and beta bands: precuneus and cingulate cortices (default mode network), middle and superior frontal gyri, angular and supramarginal gyri (fronto-parietal network), lingual gyrus (occipital-visual), lateral occipital and parietal cortices (occipital-parietal), inferior temporal gyrus and lateral occipital cortex (visual stream). Conclusion: IcEEG-fMRI allowed the investigation of the localization of alpha1, alpha2 and beta1, beta2, beta3 respectively from S1 and M1. The spatial specificity of this method allowed the identification of BOLD changes correlated with regionally specific band activity showing that alpha and beta are markers of the default mode network as well as of other resting state networks. Each frequency sub-bands was associated with a unique set of brain regions suggesting a different role played by each EEG band in the networks. This finding has implications for studies that compare RSN networks to EEG-bands and requires more investigation. 1 The Influence of Macroscopic and Microscopic Fibre Orientation Dispersion on Diffusion MR Measurements: a Monte-Carlo Simulation Study Tingting Wang, Hui Zhang, Matt G. Hall, and Daniel C. Alexander Centre for Medical Image Computing and Computer Science, University College London, London, United Kingdom Background and aims: We compare and contrast the effects of different types of fibre orientation dispersion on microstructural parameter estimates from diffusion MR. In biological tissue, fibre orientation dispersion can be divided into two classes: 1) Macroscopic fibre dispersion, a population of straight fibres with different orientation, such as in crossing and fanning structures. 2) Microscopic fibre dispersion, individual fibres with varying orientation, such as undulating fibres, which are common in nerve tissue to accommodate stretching during movement [1]. Measuring orientation dispersion is useful both for characterizing the spatial arrangement of neuronal processes but also for estimating other microstructural features, such as axon diameter and density accurately [2]. Recent parametric models of dispersion enable such estimation [3, 4, Macro Micro1 Micro2 Micro3 5]. However, these works implicitly assume macroscopic dispersion. In the presence of Fig. 1: Macro substrate (left) and Micro substrates microscopic dispersion, this assumption may bias the microstructural parameter estimates (right). Micro1, Micro2 and Micro3 are three Micro with the same ODF but decreasing [6] but the exact effects are yet to be studied. Here we construct virtual white matter substrates amplitudes. environments for each type of dispersion and conduct Monte-Carlo diffusion simulations to study differences that arise in the water dispersion, standard Diffusion Tensor Imaging (DTI) [7] indices, and parameter estimates from current biophysical models of fibre dispersion. Methods: We construct mesh fibres with biologically realistic structure and Watson Orientation Distribution Function (ODF) and then use the Monte-Carlo simulator [8] in the a b Camino diffusion MRI toolkit [9]. Macroscopic dispersion substrates contain straight cylinders with orientations nk, k=1, …, K drawn from a Watson distribution. Microscopic dispersion substrates consist of a single undulating cylinder assembled by joining straight segments with orientations n1,…nK. We control the dispersion scale of Micro substrates by varying segment c length l, and the dispersion amplitude by increasing the likelihood that consecutive segments have similar orientation. We test three Micro substrates, Micro1, Micro2 and Micro3, with the same Watson orientation distribution but decreasing amplitude. Fig.1 shows substrates c for one set of nk from a Watson distribution with dispersion parameter ĸ = 4. All substrates have axon diameter a = 6µm. We synthesize data from the ActiveAx stimulated echo Fig. 2: Estimate of (a) FA and (b) Watson dispersion acquisition protocol in Alexander et al. [10] which contains 3 HARDI shells: b-value ∈ {14631, 2306, 3425} s/mm , gradient strength ∈ {260.4, 300.00, 113.50} mT/m and diffusion time ∈ 2 {150.64, 16.33, 150.14} ms. We use the b = 3425 s/mm shell alone for DTI. The simulation 4 2 uses only intra-cellular spins with diffusivity 6x10 mm /s. We fit the Delta and Watson dispersion models in Zhang et al. [2] to estimate fibre diameter and dispersion. Experiments and results: Fig.2 plots parameter estimates from Macro and each Micro with l 2 = 4µm and ĸ ∈ {3, 4, 5, 6}. The error bars show variation over 10 different random draws of the nk. For Macro we recover the ground truth ĸ (Fig.2b) correctly, whereas we overestimate ĸ for microscopic dispersion and the overestimation increases as the dispersion amplitude decreases. This increase in anisotropy is also reflected by the increase in FA (Fig.2a). As Zhang et al. [2] shows previously, the Delta model overestimates a (Fig.2c); the bias is largest for Macro, and smallest for Micro3. Fig.3 plots estimated a and ĸ for Micro substrates with ĸ = 4 parameter, for Macro and Micro. (c) plots the estimated diameter from Watson and Delta model (should be about 6µm). a RMSD b and l ∈ {4, 10, 30, 50, 100} µm. Overestimation of a (Fig.3a) with the Delta model increases Fig. 3: Recovery of (a) a and (b) ĸ for Micro substrates with with l and is largest in Micro1. The Watson model more accurately recovers a in all cases, but ĸ = 4 and a varying segment length l, showing the trends overestimates ĸ (Fig.3b) for short l, with largest bias for Micro3. The bias gradually decreases from microscopic to macroscopic dispersion. as l increases, and the estimation converges to ground truth value when l ≥ 30 µm, at about the root-mean squared displacement (RMSD). Discussion and conclusions: We demonstrate that different types of fibre dispersion affect parameters derived from diffusion MR reconstruction techniques differently. At large scale, i.e. l at or above RMSD, microscopic dispersion produces diffusion behaviour similar to macroscopic a b dispersion, but at small scale it appears more anisotropic. Fig.4 confirms this trend. The Fig. 4: Spin displacement distribution of (a) small transition in behaviour occurs around l = RMSD, because above this value, spins rarely exchange scale Micro l = 4 µm and (b) large scale Micro l = 100µm. (a) shows greater concentration than (b). between segments with different orientations, thus diffusion ODF reflects fibre ODF; whereas below, they exchange often, diffusion ODF does not reflect fibre ODF. The parameter ĸ expresses the fibre ODF, which is important in tractography. Our results suggest that, using current methods, the ODF can be recovered accurately in macroscopic dispersion, but not with small scale microscopic dispersion. We suggest considering such effects in future modelling works. Here we test only one acquisition protocol, but the choice of diffusion times is important to define the scale and transition point in behaviour. References: 1. Fontana Florence1781. 2. Zhang et al. NIMG11. 3. Kaden et al. NIMG07. 4. Sotiropoulos et al. NIMG12. 5. Zhang et al. NIMG12. 6. Nilsson et al. NMRBiomed12. 7. Basser et al. BiopJ94. 8. Hall et al. TMI09. 9. Cook et al. ISMRM06. 10. Alexander et al. ISMRM12. META-ANALYSIS OF THE FUNCTIONAL CORRELATES OF FRONTO-PARIETAL NETWORKS Parlatini V., Catani M., Thiebaut de Schotten M. Natbrainlab, Department of Forensic and Neurodevelopmental Sciences, Institute of Psychiatry, King's College London, UK. Inserm U975; UPMC-Paris6, UMR_S 975; CNRS UMR 7225, Centre de Recherche de l'Institut du Cerveau et de la Moelle épinière, Groupe Hospitalier Pitié-Salpêtrière,75013 Paris, France Spherical deconvolution tractography has recently demonstrated that in humans, similarly to monkey, the superior longitudinal fasciculus (SLF) is composed of three distinct branches connecting parietal and frontal regions. However, their different functional roles are still largely unknown. In order to clarify this aspect, 491 fMRI studies published in the last 10 years were collected to perform coordinate-based meta-analyses of possible functions associated with SLF. Thirteen functions were analysed and included top-down and bottom-up visuo-spatial attention, spatial and verbal working memory (WM), semantic and phonological processing, saccades, motor sequences, mental imagery, emotion processing, decision making, number manipulation and functions related to the mirror neuron system. The meta-analyses were performed using the Signed Differential Mapping software and the sites of consistent activation were compared with the cortical projections of the SLF branches. Finally, paired t-tests, comparing the mean number of coordinates in the two hemispheres, were performed to evaluate the pattern of lateralisation. Our meta-analyses suggest that dorsal fronto-parietal areas connected by the SLF I contribute to voluntary visuospatial attention, oculomotor coordination, mental imagery, spatial WM and motor sequencing. Ventral fronto-parietal areas connected by the SLF III mediates language-related functions, such as verbal WM, phonological and semantic processing, but also other cognitive functions, including decision making and emotion processing. Finally, the SLF II seems to contribute to functions associated with both dorsal and ventral fronto-parietal networks and is involved in number manipulation. A clear lateralisation is observed for language (leftward) and number manipulation (righward). These findings suggest distinct functions for the SLF I and III, with SLF II contributing to both dorsal and ventral fronto-parietal networks. BC-ISMRM XXII Postgraduate Symposium 2013 Abstract details We thought it would be interesting to show you how the submitted abstracts broke down into different categories. Our sorting was necessarily quite general, but hopefully this gives you an idea of the type of work submitted for consideration. Methods 14 12 10 8 6 4 2 0 ASL DW FMRI Other Organ 25 20 15 10 5 0 Abdomen Brain Breast CNS Heart Liver Clinical / Preclinical 30 25 20 15 10 5 0 Clinical 70 PreClinical NA List of Participants At this year’s Symposium, we had nearly 140 registrants from 17 different universities and institutions. The names listed include those who agreed to have their details printed here. Dr. Nicola Ainsworth nikisimpson@doctors.org.uk University of Cambridge Adnan Alahmadi adnan.alahmadi.11@ucl.ac.uk UCL Dr. Anita Banerji anita.banerji@manchester.ac.uk Imaging Sciences Miss Claire Barnes CMBarnes@live.co.uk Cardiff university Prof. Jimmy Bell jimmy.bell@csc.mrc.ac.uk Imperial College London Dr. Mounia Beloueche-Babari mouniab@icr.ac.uk CRUK & EPSRC Cancer Imaging Centre, Institute of Cancer Research Miss Eleanor Berry eberry@fmrib.ox.ac.uk Clinical Neurosciences/Oxford Mr. Guido Buonincontri gb396@cam.ac.uk University of Cambridge Dr. David Carmichael d.carmichael@ucl.ac.uk UCL Institute of Child Health Dr. Manil Chouhan m.chouhan@ucl.ac.uk Centre for Advanced Biomedical Imaging, University College London. Mr. Muhammad Chowdhury ppxmc@nottingham.ac.uk SPMMRC, School of Physics & Astronomy, University of Nottingham Miss Isabel Christie i.christie@ucl.ac.uk University College London Matthew Cronin ppxmjc@nottingham.ac.uk School of Physics and Astronomy,The University of Nottingham Mrs. Naomi Douglas naomi.douglas@icr.ac.uk Radiotherapy and Imaging, Institute of Cancer Research Mr. Ben Duffy ben.duffy.09@ucl.ac.uk Medicine/University College London Miss Eleanor Evans ee244@cam.ac.uk Wolfson Brain Imaging Centre, Cambridge University Dr. Thomas Eykyn thomas.eykyn@kcl.ac.uk Imaging Sciences - Kings College London Dr. Pedro Ferreira p.ferreira@rbht.nhs.uk BRU, Royal Brompton Hospital Miss Dimitra Flouri mm08df@leeds.ac.uk Department of Applied Mathematics - Medical Physics/ University of Leeds Prof. David Gadian d.gadian@ucl.ac.uk UCL Institute of Child Health Miss Rupinder Ghatrora r.ghatrora@ucl.ac.uk UCL Mr. Andreas Glatz a.glatz@sms.ed.ac.uk BRIC, University of Edinburgh Prof. Xavier Golay x.golay@ucl.ac.uk UCL Institute of Neurology Mr. Miguel Goncalves miguel.goncalves.10@ucl.ac.uk UCL Mr. Francesco Grussu francesco.grussu.12@ucl.ac.uk Institute of Neurology, University College London Dr. Josef Habib j.habib@imperial.ac.uk Imperial College London Miss Hannah Hare hhare@fmrib.ox.ac.uk Clinical Neurosciences/Oxford 71 BC-ISMRM XXII Postgraduate Symposium 2013 Miss Holly Holmes h.holmes.11@ucl.ac.uk CABI, UCL Ms. Henrietta Howells henrietta.howells@kcl.ac.uk Forensic and Neurodevelopmental Sciences, Institute of Psychiatry, King's College London Ms. Andrada Ianus andrada.ianus@gmail.com CMIC/UCL Mr. Ozama Ismail o.ismail@ucl.ac.uk University College London Dr. Yann Jamin yann.jamin@icr.ac.uk Institute of Cancer Research Dr. neil jerome neil.jerome@icr.ac.uk Institute of Cancer Research Mr. Sean Peter Johnson peter.johnson@cancer.ucl.ac.uk Cancer Institute, University College London Prof. Derek Jones Jonesd27@cf.ac.uk CUBRIC, Cardiff University Dr. Tammy Kalber t.kalber@ucl.ac.uk University College London Dr. Evangelia Kaza Evangelia.Kaza@icr.ac.uk Institute of Cancer Research Mr. Grzegorz Kowalik grzegorz.kowalik.09@ucl.ac.uk UCL Centre for Cardiovascular Imaging Mr. Peter Lally peter.lally@ucl.ac.uk Institute for Women's Health, University College London Ms. Ilona Lipp lippi@cf.ac.uk CUBRIC, Cardiff University Prof. Mark Lythgoe m.lythgoe@ucl.ac.uk University College London Mr. Da Ma d.ma.11@ucl.ac.uk CABI, UCL Mr. Zaid Mahbub ppxzm@nottingham.ac.uk School of Physics & Astronomy/University of Nottingham Dr. Laura Mancini laura.mancini@ucl.nhs.uk UCL - Institutte of Neurology Prof. Ian Marshall ian.marshall@ed.ac.uk University of Edinburgh Mr. James Meakin jmeakin@fmrib.ox.ac.uk FMRIB, University of Oxford Mr. Mark Mikkelsen MikkelsenM@cardiff.ac.uk CUBRIC, School of Psychology, Cardiff University Miss Francesca Norris f.norris@ucl.ac.uk UCL Mr. James O'Callaghan james.ocallaghan.10@ucl.ac.uk CABI, UCL Ms. Ruth Oliver ruth.oliver.10@ucl.ac.uk UCL Mr. Konstantinos Papoutsis konstantinos.papoutsis@stcatz.ox.ac.uk University of Oxford Mr. Christopher Parker christopher.parker.10@ucl.ac.uk UCL Centre for Medical Image Computing Dr. Harry Parkes harry.parkes@icr.ac.uk Institute of Cancer Research Dr. Valeria Parlatini valeria.parlatini@kcl.ac.uk Institute of Psychiatry,King's College London Dr. Shreela Pauliah s.pauliah@ucl.ac.uk UCL Mr. David Peat ppxdp1@nottingham.ac.uk University of Nottingham Ms. Suejen Perani s.perani@ucl.ac.uk ICH, UCL Dr. Iain Pierce i.pierce@rbht.nhs.uk NHLI / Imperial College London Mr. Nick Powell nicholas.powell.11@ucl.ac.uk CABI, UCL Mr. Joanthan Price jonathan.price@icr.ac.uk Institute of Cancer Research Mr. Alessandro Proverbio alessandro.proverbio.09@ucl.ac.uk University College London Mr. Rajiv Ramasawmy rmaprra@live.ucl.ac.uk Centre for Advanced Biomedical Imaging Miss Marilena Rega marilena.rega.10@ucl.ac.uk university college london 72 Mr. Simon Richardson simon.richardson.09@ucl.ac.uk CABI Mr. Tom Roberts thomas.roberts.10@ucl.ac.uk CABI UCL Mr. James Ross james.ross@abdn.ac.uk University of Aberdeen Mr. Stefano Sandrone sandrone.stefano@hsr.it Vita-Salute San Raffaele University Dr. Stephen Sawiak sjs80@cam.ac.uk University of Cambridge Dr. Karin Shmueli k.shmueli@ucl.ac.uk Department of Medical Physics and Bioengineering, UCL Dr. Nour Shublaq n.shublaq@ucl.ac.uk Centre for Computational Science, UCL Mr. Robin Simpson rmsimpson175@gmail.com NHLI, Imperial College, London Dr. Bernard Siow b.siow@ucl.ac.uk CABI, CMIC / UCL Dr. Po-Wah So po-wah.so@kcl.ac.uk King's College London Ms. Jessica Steventon steventonjj@cardiff.ac.uk CUBRIC, Cardiff University Mr. Alan Stone stonea9@cardiff.ac.uk CUBRIC, School of Psychology, Cardiff University Dr. Daniel Stuckey d.stuckey@ucl.ac.uk CABI/UCL Dr. Magdalena Szafraniec magdalena.szafraniec@gmail.com University College London Dr. David Thomas d.thomas@ucl.ac.uk UCL Institute of Neurology Mr. Patxi Torrealdea papomail@gmail.com UCL Institute of Neurology Dr. Anthony Vernon anthony.vernon@kcl.ac.uk King's College London Dr. Simon WalkerSamuel simon.walkersamuel@ucl.ac.uk UCL Centre for Advanced Biomedical Imaging Miss Tingting Wang tingting.wang.11@ucl.ac.uk University College London Miss Esther Warnert warnertea@cardiff.ac.uk CUBRIC Jack Wells jack.wells@ucl.ac.uk UCL Claudia WheelerKingshott c.wheeler-kingshott@ucl.ac.uk University College London Emily Wholey emily.wholey@icr.ac.uk Institute of Cancer Research Mr. Thomas Wilkinson thomas.wilkinson2@postgrad.manchester.ac.uk University of Manchester Dr. Jessica Winfield jessica.winfield@icr.ac.uk Institute of Cancer Research Dr. Gavin Winston g.winston@ucl.ac.uk UCL Institute of Neurology Yichao Yu ucbtyyu@ucl.ac.uk UCL Dr. 73 BC-ISMRM XXII Postgraduate Symposium 2013 Notes 74 Notes 75 BC-ISMRM XXII Postgraduate Symposium 2013 Notes 76 Local Area Map Programme 9:00 Williams Lounge Registration opens 9:30 Henry Wellcome Auditorium Welcome & Opening remarks Henry Wellcome Auditorium Oral presentations: Cutting-edge techniques 11:15 Williams Lounge Refreshments 11:45 Henry Wellcome Auditorium Oral presentations: Brain Investigations 9:35 13:30 Prof. Mark Lythgoe Chair: Prof. Steven Williams Chair: Prof. Derek Jones Dr. Ken Arnold (Head of Public Programmes, Wellcome Collection) will introduce the Collection just before lunch. 13:35 Williams Lounge Lunch & Exploring Wellcome Collection 15:00 Henry Wellcome Auditorium Poster Pitches 16:30 Franks & Steel Rooms Posters & Refreshments 17:30 Henry Wellcome Auditorium Oral presentations: Cancer, Cardiac & Nerves Henry Wellcome Auditorium Prizes and Concluding remarks 18:30 19:00 Chair: Prof. Mark Lythgoe Chair: Risto Kauppinen Prof. David Gadian and Prof. Mark Lythgoe Wellcome Collection closes

of the International Society for Magnetic Resonance in Medicine

Related documents

Products

Support

of the International Society for Magnetic Resonance in Medicine

Related documents

Add this document to collection(s)

Add this document to saved

Suggest us how to improve StudyLib